Last Sunday, a researcher and fellow Substack writer Citrini posted a question on X. It began in the future tense: June 2028. The S&P had plunged 38% from its peak. Unemployment clocked in at 10.2%. AI had not only met expectations, it had shattered them.

What happened?

The thread linked to a full article. Twelve thousand words of speculative macro fiction of AI agents replacing half the white-collar workforce, private credit unraveling, prime mortgages cracking in tech hubs, a deflationary spiral caused by how good the technology is by wiping out need for a bunch of excess labor.

It was a thought exercise, from the future. Citrini said.

Twenty-one million people read the post, and forty-four thousand bookmarked it. Investors around the world copy and paste the post into their own corporate-sponsored ChatGPT and ask: Stress test my positions against this thesis, assuming it’s ALL TRUE. Make no mistakes.

When markets opened Monday morning, DoorDash tumbled 8%. American Express followed suit. Apollo, Blackstone, Visa, Mastercard, ServiceNow. Every company named in the thought experiment was affected. The iShares software ETF hovered near a 52-week low, wiping out nearly all the gains since ChatGPT’s debut in November 2022. Three years of the AI trade, erased before lunch. Bloomberg and The Wall Street Journal both ran with the story, pinning the selloff on a Substack post. All this, triggered by a piece of fiction, openly labeled as such, penned by a researcher unknown to most of Wall Street.

Ten months earlier. An account called Walter Bloomberg (not affiliated with Bloomberg), just a dude with a blue checkmark and a propensity of posting headlines in all caps, writes: HASSETT: TRUMP IS CONSIDERING A 90-DAY PAUSE IN TARIFFS FOR ALL COUNTRIES EXCEPT CHINA.

It was fake news. But the headline had already escaped. CNBC read it live and Reuters published it. In just ten minutes, the S&P soared 8%. Two point four trillion dollars moved. The White House made its disclaimer and Walter Bloomberg deleted the post.

Two days later, Trump actually paused the tariffs for ninety days, exactly as the fake headline had described. But hours before announcing the real pause , the one that would add four trillion dollars back to the market, he posted on Truth Social, also in all caps: THIS IS A GREAT TIME TO BUY!!! DJT.

And then October. A single Truth Social post, “massive increase” in tariffs on China, erased more than hundreds of billions from U.S. semi stocks market cap in a day. Trade deals were announced, tweaked, scrapped, and revived, all through social media, all at the speed of a keystroke, long before the machinery of government could even open a Word document to draft a response.

What once demanded a Soros is now open to any theorycel with a Substack. A researcher posts a scenario and the scenario shapes the sector, at least the perception of it. A blue checkmark posts a headline and the headline becomes policy. Soros’ reflexive loop is now compressed, democratized, stripped of the institutional drag that once gave people time to think. We’ve evolved past his framework. The loop no longer needs to be understood to work. It no longer requires meaning. It only needs motion.

This is headline economics, and its logic recursive.

The Carcass

Yorgos Lanthimos’s new film Bugonia tells the story of a kidnapping that also serves as a test of what we know and believe. A man kidnaps a female CEO because he thinks her company is killing the bees and his mother, and that she is an alien following orders from an emperor in Andromeda. He turns out to be right about everything. Still, Lanthimos, who might be the cruelest filmmaker working today, makes sure that this truth does not change anything.

The conspiracy theorist has the facts. His accuracy is beside the point. What his pain requires is not truth, but a container: a villain, a structure, an address to which suffering can be sent. Without one, pain is ambient and therefore permanent. Conspiracy theory is compensatory agency, by restoring a sense of participation in a system that has locked one out of every lever.

Emma Stone, the CEO, inhabits the opposite epistemology. She doesn’t need to make up any new story but simply exist in the perpetuating logic of capitalism. “My company is a key job creator and economic engine of the region. I am important,” she said with sincerity. There’s no irony because her claim is coherent to her worldview. She can’t see what he sees, because her instruments were calibrated for a different set of heuristics entirely.

The film’s title comes from Virgil. Bugonia is the old belief that bees are born from the body of a sacrificed ox. It is the idea of life coming from decay, sweetness from death. In truth, what actually came out of the ox were drone flies, colored black and gold, fast enough to look like honey bees. People want to believe so badly, because facing the truth (that the ox died for nothing, that sweetness will never come, that decay is just decay) is uncomfortable.

Which is to say, people see what their position requires. And in America, position has split clean in two. The K-shaped split in the economy doesn’t just bifurcate economic outcomes, but perpetuate different narratives people tell themselves. The needs from either end are so different that they create worldviews that can’t be reconciled, each making sense on its own but impossible for the other side to see.

The conspiracy theorist’s paranoia creates a villain and a story that keeps his fears going. The executive’s power creates the incentives that support her position. The contrast between these two ways of thinkings represent a defining characteristics of markets.

Soros came close to describing this through his seminal work in Reflexivity, and Schiller’s narrative economics brought us even closer. And neither fully captured what happens once latency tends to zero. Soros’s reflexive loop implies that the cognitive function has time to update when presented net new information. And Schiller’s narratives spread like epidemics, which require time for incubation. In headline economics, contagion detonates everywhere all at once. Reflexivity arrives as an explosion before we have time to understand what’s going on.

When latency approaches zero

In Narrative Economics, Shiller argued that stories spread like epidemics and move markets the way viruses move through populations. The dotcom boom, the subprime bubble, crypto. Narratives helped create these bubbles. They shaped economic behavior, making people buy, panic, or hold out based on stories rather than clear reasoning.

Shiller’s narratives spread only as fast as people could talk. Dinner parties, newspaper columns, a coworker mentioning his brother-in-law who made money flipping houses in Phoenix. The housing story played out over ten years, slowly unraveling, as documented in The Big Short. Bitcoin took years, too. People had time to think about information and decide what to do, which now feels almost quaint.

The American lawmaking system was an additional buffer, giving markets time to determine prices. Policy went through hearings, markups, long stretches of floor votes. Markets could process these signals, but there were still physical limits that required people to work together. Now, the policy (or potential policy) shows up as a short, capitalized message on the screen. The gap between when something happens and when it is verified has widened. Policies don’t have to go through the bureaucracy to become a signal.

The headline describes what happened and simultaneously becomes the event itself.

This is what headline economics shows: markets trade based on how easily a claim can be turned into a position, how clear its direction is, and how widely it spreads. Tradability is the main feature. What matters most is whether the headline’s effect on price is big enough to turn voluntary trades into forced ones. Once the price crosses a certain threshold, algorithms take over. In a zero-latency market, forced liquidations happen first. The basic reality of an event now matters less than its second derivative: the fast-moving momentum of structural capitulation.

Citrini’s story became powerful because it was so easy to understand. It did not rely on vague fear, but clearly named the casualties, removing the need for interpretation by serving ready-made pair trades on a plate. One doesn’t need to ask if something is real, but whether they can act on the incoming momentum before someone else does.

Soros called this reflexivity, meaning the market watches itself and changes what it sees in the process. The market does not just reflect reality; it rewrites the story as it goes, like a marriage that changes through the acts of faith or a child who behaves differently when being watched by their parents.

Stripped of context and propagated at network speed, the headline is now a unit of economic reality, among many. Its fragmentation and speed grant it power through algorithms, allowing it to travel faster than verification.

The bruise and the flinch

At the start of Bugonia, you already sense Teddy’s tension even before you know why. You see it in Plemons’s posture as he rides the bus to his warehouse job at Auxolith, clocking in for a company he’s secretly working against. Emma Stone’s CEO is simply the face of his fantasy.

The Citrini post didn’t move billions because it offered any groundbreaking insight. In fact, most fund managers had already imagined this scenario, likely during late nights spent worrying about their worst fears. The thought that AI could succeed but still lead to disaster was already on their minds, lingering in their portfolios and sleepless pre-market hours, waiting for something to make it real. Citrini did that.

You can’t trade on a feeling of vague dread. But once that fear becomes clear, it can move markets. The post simply set up a tension that’s already there.

A tense market is a supersaturated solution. It has already absorbed more than it can hold; it’s just waiting for the grain. That’s all a headline is, on the days it matters, and most days it doesn’t, most days it’s a stone in still water. It presses where it hurts, and the flinch tells you how deep the bruise goes.

Although a headline can push prices out of line, but the dislocation rarely lasts. The S&P bottomed in March 2020 and was back at pre-pandemic levels within months even before the vaccines and reopening. The tariff scares resolved the same way, where in a few weeks, the price found its way back to its value. This is usually told as a story about resilience, but recovery is not the same as restoration. Something gets spent in the round trip, and what gets broken doesn’t show up in the index.

In a leveraged market, just twelve minutes of incorrect prices can cause lasting harm. Stop-loss orders triggered during Walter Bloomberg’s twelve minutes led to real losses based on a signal that never existed. A margin call caused by false information does not get undone when the truth comes out. The retail investor who sold her 401(k) near the bottom, unable to watch her balance drop any further, was already out by the time the market rebounded in August. She missed the recovery. The Salesforce product manager who lost her $180,000-a-year job during the panic will not get a call back when things calm down and the index turns green again. Now, she drives for Uber.

The price recovered, but her life did not.

K-shaped economy and K-shaped perception

The damage does not fall equally.

In what’s called a K-shaped economy, the bottom half of Americans—about 165 million people—own just 2.5% of the nation’s wealth. Meanwhile, the top 10% controls 87% of the stock market. Since ChatGPT came out, AI stocks have made up three quarters of the S&P’s returns. When people talk about the economy’s performance, it’s really just a small group benefiting while most of the country is left out.

The fund manager who sells ServiceNow in a panic on Monday morning is the same one who buys Micron later that day. The Citrini post was negative about software and payments, but extremely optimistic about AI infrastructure—if you could read it that way.

At the top of the K, where people have the power to move markets, even reading a headline is like making a trade. The Citrini post sparks fear, which leads to selling. That selling becomes a Bloomberg headline, which causes even more fear and more selling. Thinking and acting happen almost at the same time. Portfolios are rebalanced before the second cup of coffee is finished.

At the bottom of the K, all you can do is think and worry. You read the same post and feel the same dread. Maybe you’re a Salesforce product manager making $180k who gets laid off and ends up driving Uber for $45k. Maybe you’re a 760-FICO borrower in SF who did everything right and are watching the ground shift under a mortgage that was good on the day it was written. You feel all of it. The emotional amplitude is identical to the fund manager’s. The economic amplitude is a rounding error.

You can’t make a move. Your RSUs haven’t vested, or they’re so far underwater that you feel stuck. You can’t hedge your bets. You can’t short the stocks you know are in trouble. You can’t switch into Micron because you don’t even own it. What you do have is a mortgage. Your fear has nowhere to go except sharing the post, bookmarking it, arguing online, or texting a friend, “we’re SO fucked!!!!”

We know that every compression of a reflexive cycle eventually forced something new. The South Sea Bubble created the first financial regulation. In 1929, the SEC, and in 08, Dodd-Frank. Each time, the cycle kept spinning until it broke; and then we rebuild.

Tomorrow morning, there will be another headline. Numbers move before the first sip of coffee; the dread gets a ticker and now a direction, and for a second that almost feels like relief, because at least now it has a name.

“Civilization advances by extending the number of important operations which we can perform without thinking of them.”

— Alfred North Whitehead

I.

The café on Fillmore buzzed with noise. I saw a little girl, about six or seven, leaning over the table, lost in a math problem. She tapped her fingers in small groups and moved her lips as she worked. Her mother watched for a bit, then leaned in to help.

“Honey, remember? We memorized it,” the mother said calmly, but she sounded tired. “You don’t have to think about it. You know it.”

In January 2026, DeepSeek released a paper that effectively said the same thing to the most powerful AI on earth.

The paper, Conditional Memory via Scalable Lookup, makes a point that is technical as it is philosophical. The Transformer, the basis of modern AI, cannot distinguish between thinking and remembering. We use it like a universal computer, making it reconstruct information from scratch every time it answers.

For example, if you ask about “Diana, Princess of Wales,” the model doesn’t just pull up the answer. Instead, it rebuilds the idea of Diana through its own neural network, step by step, until it gets to the answer.

DeepSeek calls this “an expensive runtime reconstruction of a static lookup table.”

This approach is analogous to that of the little girl in the café, working out fifty-six by counting on her fingers. It means thinking through something every time, even when the answer should be easy to recall.

II.

The struggle between solving problems and finding information has existed for ages. This back-and-forth goes all the way back to the clay tablets of Uruk.

Babylonian astronomers made Lookup Tables so they wouldn’t have to calculate the positions of the planets every night. Doing the math was hard work and meant scribes had to stay up late, using oil lamps. Instead, they wrote the answers on clay tablets. If you wanted to know where Venus would be, you just checked the library instead of doing the math yourself. In this way, they used more space for clay tablets to save time on calculations.

This trade-off has shaped the entire history of computing.

In the 1940s, memory was hard to come by, so computers stored bits as sound waves in mercury tubes. Because memory was limited, machines had to redo calculations they couldn’t store. By the 1970s, cheaper RAM changed things. Now, computers can save results and avoid repeating work. In the 1990s, processors became so fast that they often had to wait for memory. To fix this, engineers added a small, fast cache to keep important data close to the processor.

Large language models made us depend on raw computing power more than anyone expected. Through the Bitter Lesson, we found out that handcrafted knowledge could not match brute-force methods that simply churn through a lot of compute. Instead of giving models instructions, we let them learn everything from the ground up, mixing facts and reasoning in one system. This method worked, even though for a while, nobody really understood why. But there was a catch: every fact, even simple ones, had to be recreated by the whole network. Now, we use a supercomputer just to check if Paris is in France.

Engineers in China, operating under computational constraints not encountered by their American counterparts, experienced these inefficiencies most directly. Once again, constraints inspire creativity. DeepSeek’s solution, Engram, represents a modern adaptation of the N-gram lookup method, previously considered obsolete, reimagined as a rapid and scalable memory system. This approach enables the model to recognize familiar patterns and retrieve them directly, rather than recomputing, effectively introducing a cache mechanism to transformer architectures.

It allows the model to say: I know this pattern. I don’t need to think about it. I’ll just fetch it.

III.

To look away from the machine and towards the classroom is to see the technical error restated as a crisis of upbringing. We have presided over a bifurcation of the intellect, dividing the mind into two camps in its effort to understand the architecture of the mind.

In the West, especially in elite American education, memory is often viewed with suspicion. Rote memory is seen as a weak substitute for genuine understanding. What matters is critical thinking: expecting children to form opinions grounded in strong logic. The strength of this approach is undeniable. It creates a “self” that is resilient against the unknown, capable of navigating terrain where no map exists. It produces the kind of elasticity that allows a mind to refuse the status quo and invent a new one.

Now, picture the world of the Gaokao, Hagwon, or a Kumon center. A child’s hours after school stretch as long as the school day itself; a single number delineating the shape of the entire life course.

The intensive drill-based educational system in Asia can look analogous to LLM pre-training, in which memory and the illusion of understanding can be solidified through repetition. These systems rely on procedural memory and instinctual recall. In a way, fluency functions as an extensive internalized reference system.

Taken to their logical extremes, both systems fail the very students they claim to serve, yielding not mastery but distinct forms of mental stagnation.

A primary limitation of the ‘Eastern’ educational model is overfitting. Students may solve a test problem with mechanical precision, but struggle when confronted with a novel variation. This approach produces learners who become static repositories of information, rich in procedural knowledge but lacking in adaptive reasoning. Such outcomes support critiques that, while these systems excel in efficiency and scale, they often inhibit creativity. This phenomenon appears to be a structural consequence of excessive reliance on memorization rather than a purely cultural issue.

The “Western” model, on the other hand, produces a student of dazzling conceptual agility, who can eloquently explain why multiplication works and deconstruct the colonial history of justice, but who cannot calculate the tip at a dinner party without panic. They are CPUs that have overheated. They have been taught to think so critically that they have forgotten how to know.

This is why the DeepSeek paper can be so edifying. When engineers let the model memorize simple, everyday facts, performance on reasoning tasks improved meaningfully. The leap in reasoning outpaced even the gains in knowledge.

These results highlight a limitation often neglected by prevailing educational philosophies: reasoning capacity is finite, analogous to the context window of a computational model. Cognitive resources are limited, and time spent on basic problem-solving detracts from creative or original intellectual work.

When a pianist plays Rachmaninoff, they don’t think about every single note as they play; the piece instead flows through their fingertips like a well-timed storm. The performance is unhindered by technicalities. If the pianist had to think about finger placement for each note, like a student solving a problem from scratch, the music would lose its flow. By memorizing the basics, the mind is free to focus on expression.

What if memorization does not hold us back, but actually helps us think more freely?

IV.

In studying cognitive processes, the researchers at DeepSeek have identified the “U-shaped curve” to determine the optimal allocation of neural resources between the “Engram,” the fixed memory store, and the “Experts,” responsible for active reasoning.

Both extremes can be unforgiving. If too little memory is used, the system can’t develop and stays stuck, unable to remember things, like a child who can’t learn. But if too much memory is used, the system becomes rigid and stops reasoning, relying solely on what existed before.

The optimal configuration identified was an asymmetrical distribution: 25% allocated to memory and 75% to active reasoning.

When memory allocation exceeds this threshold (typically called a Hash Collision), overfitting occurs. Encountering the word “Bank” may prompt the immediate recall of a riverbank, even when the context pertains to a financial institution. This misretrieval exemplifies the challenges experts face, who often rely on experience rather than context-specific reasoning.

Everyone knows the senior partner at a law firm who has seen everything. He sits behind glass, radiating experience, cutting you off before you finish. Usually, he already knows your question. He operates like a living lookup table. Nothing seems to make him flinch. He spots the pattern, recalls what worked in 1998, and applies it with confidence.

He moves with speed and authority; however, his reliance on outdated knowledge can result in errors. The accumulation of past experiences may hinder his ability to accurately assess current situations, as excessive recall diminishes critical inquiry.

It’s a huge efficiency gain that so much human knowledge can be accessed simply by recalling it, rather than working it out each time. But if we relied on memory without any checks, we could easily hallucinate. That’s why we need something like a Gate between stored memories and active reasoning. This gate does more than just open or close. It looks at what we remember, catches easy answers or familiar solutions, and tests them before information passes through.

The Gate helps us distinguish between situations. It stops us from running on autopilot. It reminds us: even if this crisis looks like the Dot Com bubble and the chart looks like 1999, the details are different. Don’t rely on memory. Don’t just rely on pattern recognition. Think it through.

V.

The DeepSeek paper asks a question that seems to be about computer chips, but it’s really about us. It’s a question of how we use our limited attention.

With only so much attention to give at any moment, how should we use it? How much should go to the parts of us that solve problems as they come up, and how much to the parts that simply remember what we already know?

Most of us are living far out on the edges of the U-shaped curve, suffering the high validation loss of a misallocated life.

I have friends who are under-cached. Every email is a fresh existential crisis, every menu a small referendum on who they are. They would tell you they guide themselves with first principles, authentic, a pure MoE machine, running hot, believing friction to be virtue. And they are exhausted. By refusing to automate the syntax of their lives, they have nothing left for spontaneity or poetry, Nietzsche’s Dionysian, the animal spirits.

The over-cached are worse, potentially, delegating way too generously to buzzwords, borrowed heuristics, an anecdote for everything and an insight for nothing. Memory scaled to the billions while the reasoning starves. Their Gate quit years ago. They aren’t living so much as re-running days they already survived, and the rerun gets cheaper every season.

Twenty-five percent less thinking is, paradoxically, a heuristic one can borrow. It’s not up to any mortal to allocate our tokens with the precision of a model architecture, but we can be deliberate about what we let harden in our cache.

So it’s time to compile. The multiplication tables, the polite greetings, the commute, the filing of tasks into their cabinets, all of it pushed down into the Engram where it runs without me. What’s left over, I spend at the edges: the heartbreak, a blank page, a glance from a stranger.

When the little girl in the café one day blurts out “fifty-six” without her fingers and looks up, she starts noticing the world around her. Then we’re reminded once again that memorization is not an escape from thought. We remember just enough so that, at last, real thinking can start.

Happy New Year, explorers.

Each year, I pose a set of questions rather than predictions. The goal is not to forecast the future, but to notice what’s already changing beneath our feet.

This year’s questions circle a single transformation: software is becoming infrastructure. The apps and dashboards we built for human eyes are giving way to systems designed for machines talking to machines, with humans hovering at the edges, deciding, permitting, revoking, and of course, curating. The interface isn’t disappearing. It’s just no longer where most of the action is.

My investment and writing focus in the upcoming years will likely stem from these inquiries.

You can find my previous inquiries here (2024) or here (2025).

The end of UI-as-destination

For two decades, software has revolved around people, with dashboards and feeds as our main control panels. These remain vital for decisions like choosing, paying, and revoking. But the spotlight is moving to agent distribution. Once agents are empowered, they can wield capabilities at scale. Human attention still sets the pace for trust and adoption, shaping how we select, buy, permit, and build habits. Yet when it comes to real-world use, the bottleneck shifts to the agents themselves: their ability to route, pick tools, manage call budgets, and recover from errors becomes the new frontier.

So what replaces the app store in this world? Maybe a capability registry. Maybe an enterprise allowlist. Maybe just the default toolkit from whatever platform wins the initial trust game. And “brand”—a fuzzy human construct—what does it become when customers are algorithms? Perhaps a machine-readable trust score: uptime guarantees, verified attestations, dispute resolution history. The marketing layer collapses into metadata.

The war of standards

When the stack becomes modular, the focus shifts to setting standards. For example, the agentic stack is converging towards a couple (still non-consensus) standards:

• MCP (Model Context Protocol) defines how agents connect to capabilities—the handshake layer.

• A2A governs agent-to-agent coordination.

• AG-UI manages the messy interface between nondeterministic AI and deterministic humans.

Where does value pool when everything interoperates? Who decides what gets included in the default spec? What becomes the DNS for agents—the universal directory of capabilities, reputations, revocations? And critically, what’s the TLS-equivalent for intent, the cryptographic proof that an agent’s request actually represents a human’s will?

Traces as compounding moat

Static data depreciates. What compounds is instrumented traces: every attempted action, every tool error, every rollback, every “unknown unknown” turned into a test. Regulators are converging on logging and traceability requirements for higher-risk systems. Meanwhile, software has standardized on Open Telemetry nd architectures like event sourcing already treat the log as the record of truth.

The big change is that traces now help systems improve themselves. A mature agent stack does more than just notice failures; it turns them into structured reports and then into improvements, all on its own. This is where behavioral versioning becomes real. For example, “agent v3.2.1” is defined not just by its weights, but by how it acts under different rules like how it rolls back, where it refuses, how it retries, and how it handles disputes.

Will we get systems that automatically mine their own traces into regression tests, adversarial prompts, failure catalogs? Is there a universal trace format, or does every vertical need bespoke observability? And how do you prevent the classic reinforcement learning trap—the agent learning to game the metrics instead of achieving the goal?

Machine identity =! human identity

Passkeys offer secure, easy login for people, but agents need a different way to prove identity, demonstrate trustworthiness, and gain permission for specific tasks. Basic tools for agent environments exist: Firecracker microVMs separate tasks serverlessly, V8 isolates run quick edge tasks, and container sandboxes like Daytona enhance security.

What would a ‘passkey for an agent run’ look like? It could be a special hardware key linked to where the process runs, using a way to prove it is genuine from far away. How can we safely give permission, for example, letting agent A spend up to a certain amount at a store until a set time, and take back that permission even if there are problems with retries or the network? What is the computer version of ‘consent’—a permission that people can understand, computers can check, and that can be reviewed later?

When remembering becomes a vulnerability

Personalization promised convenience. Persistent memory delivered it. But now memory isn’t just a feature—it’s an attack surface, reshaping the implicit contracts between users and systems.

OpenAI’s memory feature reaches across your entire chat history, synthesizing patterns from fragments you’ve long forgotten. They offer transparency controls. But delete a conversation and ask yourself is it really gone? Or has it already influenced the model’s understanding of you in ways that can’t be cleanly excised?

What kinds of memory are we creating? Are we using logs that can only grow, making real deletion impossible? Are we spreading your patterns across data in ways that can’t be separated back into facts? Are we syncing memories across devices, making it hard to see yourself as just one person? The decisions we make in 2026 will shape what it means to “forget.”

Robotics without robots

Progress in robotics now depends more on general perception-to-action abilities and practical tools than on mechanical advances. OpenVLA, a 7-billion parameter vision-language-action model trained on 970,000 real robot episodes, shows strong general manipulation skills. DeepMind’s Gemini Robotics offers on-device VLA fine-tuning for developers. NVIDIA’s Isaac GR00T N1 is promoted as an open, customizable foundation model for humanoid robots. Software has been, and will continue to pull hardware forward.

Collecting robot data is costly and slow. Scaling up data collection, possibly through teleoperation, sim-to-real transfer, or crowdsourced demonstrations, could be crucial for improving models. Could we create standardized robots or components, similar to PC-compatible hardware, to speed up hardware updates and model training? The best idea may be interoperable, programmable parts, such as a ‘Third Eye’ vision module with built-in inference and networking, that can be added to any machine to give it robotic abilities. A company making these components, rather than full robots, would benefit each time a device uses them to see and act.

Industrialization of taste

Turns out discernment can be mass-produced. Generate hundreds of candidates, filter through learned reward models, iterate until output matches stated goal. Research on preference optimization explodes with mid-training techniques, post-training strategies, automated evaluation frameworks.

So when taste scales, where does the bottleneck shift? To objective functions. To whoever defines “good” and controls the evaluation. (We already know benchmarks are broken.) Aesthetic monoculture becomes plausible. When evaluators converge on similar proxies—coherence, novelty, “vibes”—outputs homogenize across domains. Some fields resist this. Law, therapy, strategy: domains where “good” is irreducibly contextual, where the relationship between practitioner and client matters as much as the output. Watch what breaks when we force-fit industrial taste onto these resistant domains.

Prompting beyond typing

We’re already seeing voice-to-text interface taking off on-device, with a small whisper model doing a great job. We’re still too early in how fast we can transmit our thoughts to the machine, and prompting isn’t going to look like typing on a keyboard or even speaking to a computer. It’s not fast enough. 12 people now have received BCI implants, so is that what’s coming next? Maybe before BCI arrives, we’ll start texting and speaking in a new way.

Ghost Hollywood

Latvia’s Flow won the Oscar for Best Animated Feature and was created because he was somehow obsessed Blender, the free/open-source 3D modeling tool. It’s now consensus that it’s just a matter of time that an AI-made film could be a worthy contender in mainstream awards. Creative production pipelines are increasingly “node graph native.” ComfyUI’s graph-based diffusion orchestration is an archetype of how creators build controllable, remixable workflows.

But feature-length production still hits limits, and not necessarily bounded by model quality. The reagents now are character continuity, shot matching, asset management, editorial control, and versioning. All the boring middleware. Will these get Bitter Lessoned away, or do they resist pure scaling because they encode human aesthetic preferences?

Race to escape the permanent underclass

The speculative energy of crypto and meme stocks has shifted from the early impulse of greed—gamble your way into generational wealth–into the current impulse of fear, to escape the permanent underclass that a small cadre of AGI-defining corporations might impose. Scott Alexander’s recent post argues that the best response to AI displacement isn’t optimization or doomerism—it’s play. Figure out what you actually want. Do the thing you love doing.

This is harder than it sounds. We’ve been socialized to optimize for homogeneity, not edge. But just as retail orderflow now moves institutional markets, the vernacular of trading has permeated everyday life. Everyone knows the phrase now: find your alpha.

If you’ve already internalized that this is a historically significant moment, and that most of us will be “small” relative to the forces reshaping everything, then the asymmetric bet is to do things anyway. Small actions in pivotal moments compound strangely. The person tinkering in the garage while empires consolidate has been the protagonist before.

And this leads to perhaps one of the most important inquiry of this decade: how do we get people to awaken into their power and do stuff?

"For here there is no place / that does not see you. You must change your life."

— Rilke’s Archaic Torso of Apollo

Building a large language model happens in three main stages.

Pre-training processes trillions of tokens scraped from everywhere, taken in with little discrimination. The model learns to predict the next word from the last, and the deceptively simple task demands massive compute. Tens of thousands of GPUs for months, tolerating messy, uncurated data dredged from the internet’s sediment. Each layer of data becomes another layer of silt, building a multidimensional echo of the collected human experience. The accumulated text of human civilization fed forward without editorial judgment. Here, more is more.

Post-training is the inverse: the model now exists, and the question shifts from what it knows to what it becomes. Data quality becomes everything—reinforcement learning with human feedback, direct preference optimization, constitutional AI, rejection sampling with teacher committees. A few perfect examples outweigh a million mediocre ones. Quality is compression, excellence, exclusion. Compute footprint shrinks to a fraction of pre-training’s cost. The curatorial burden expands to absorb it. Someone has to decide what ‘good’ looks like.

Mid-training sits between these phases, though fewer practitioners name it directly yet. It is the industrialization of quality, the discovery that you can manufacture discernment at scale. The model is fed carefully constructed data: machine-generated text filtered by other machines trained to judge it, high-quality sources deliberately overrepresented, answers transformed into questions so the model learns to ask what it already knows how to solve. This is neither the indiscriminate hunger of pre-training nor the precious hand-curation of post-training. It is the assembly line where taste is mass-produced. The foundation is set, but values remain unwritten. In this liminal space, the model is shown, again and again, what 'better' looks like.

LLM metaphors have begun to infiltrate my everyday language. I’ll describe someone’s intellect as ‘high tokens per second,’ request a ‘context window’ before a catch-up, or laugh about ‘updating my weights’ after criticism. Like all metaphors, these shrink the wildness of people into neat, lossy packages. Yet the three training phases persist in my mind—a powerful lens, a rough taxonomy for not just how people think, but how they navigate the world.

But 2025 complicated this triptych. The industry began to behave as if there were a new major stage, longer and hungrier than the thin epochs of instruction-tuning we had grown used to. What changed was the discovery of a specific asymmetry, the realization that while it is difficult to solve a complex problem, it is trivial to recognize the solution. A Sudoku puzzle, a line of code, a proof. These are tasks with a hard, opaque front and a transparent back.

We left behind the era of ‘taste’—that delicate, costly consensus that resists scaling—and entered the age of the indisputable result. Once rewards became verifiable and objective, the gentle, exhausting labor of human judgment was no longer required. Now, the model simply grinds against an indifferent reality, unpersuadable and unmoved, until it collides with the truth.

We once spent everything on “eating the world,” a vast ingestion followed by the decorative sculpting of human preference. But the focus has shifted. Now, the model is left to grind against tasks that possess a fixed, indifferent clarity. It invents a private ladder. And it climbs without a break.

The pre-training mind is defined by a frantic, lateral hunger. It is driven by two things: volume of data and the brute force of compute. As a result, they read everything, take every meeting, and believe it’s all “a numbers game.” We call them “machines,” a term that has become less a slur and more a kind of secular canonization. Their work ethic is undeniable, yet there is something ghostly about their trajectory; they run with a ferocity that suggests they are less interested in the destination than in the pure, frictionless sensation of the running.

Their knowledge is vast and rhizomatic, a sprawling map with no capital. They speak on any topic with a fluency that is as impressive as it is ungrounded—a glittering surface with no detectable taproot. If you ask them what it is all for, the question strikes them as a category error. Why wouldn’t you do this? For them, the exhilaration of capacity is its own justification. They are the architects of the hive, comfortable managing armies toward a goal that remains conveniently abstract.

There is a sense of collectivism here, a philosophy that metabolizes difficulty into capability without ever pausing to ask what that capability serves. Understanding is not a requirement; coordination is the only sacrament. In this regime, belonging is the operating system. When you are part of the collective, purpose doesn’t need to be articulated; it can be assumed.

By conventional metrics, they have won. They have inhaled the earth. And it is important to realize that their victory is not a relic. Critics or tinkerers did not build the massive, humming data centers of 2025; they were willed into existence by the grit of the pre-training mind. They are the only ones capable of the “more is more” logistics required to pave the planet in silicon.

But 2025 introduced a new, chilling irony into their labor. They spent their lives building the cathedral of “scale,” assuming that the height of the ceiling would be the measure of their soul. Instead, they discovered that they had merely built a very large room for a very different guest.

Reasoning is indifferent to how much you consume; it cares only for the friction of hesitation. It is the model’s secret dialogue, testing and discarding ideas in the shadows before a single word comes out. For the pre-training mind, this is an existential crisis! They have become the system’s oxygen — essential, everywhere, and invisible to the very life they uphold. They are the pavement’s strength, gazing up at the passing cars, bewildered by a world they no longer know how to steer.

The post-training minds are rare, and they know it. They cultivate their scarcity as an identity. These are the people who read primary sources, who spend an afternoon at the Frick, who refuse meetings with devastating politeness. It isn’t just snobbery; it’s the terror that their attention, the only thing they truly own, is being diluted by the world’s noise. They care about things that feel like a nuisance to everyone else: the grain of a specific paper, the historical baggage of a typeface, the minute, agonizing space between two words that a machine would treat as identical.

Something dishonest sits at the center of their discernment. They rarely admit, even to themselves, that their refined taste was trained on a corpus they did not personally ingest. They inherited a foundation and called it “vision.” They stand on the roof of a house and critique the masonry of the basement, forgetting they didn’t carry a single brick. For them, every choice is a debt: the burden of making something worthy of the accumulation they rely on.

They see the shape of excellence so early and so vividly that the act of beginning becomes a betrayal. The gap between the vision and the execution is not a distance to be traveled, but a canyon that makes the first step look absurd. To do nothing begins to feel like a moral act—a refusal to further pollute the world with the very slop they have spent their lives learning to despise. They are the world’s most gifted editors, which is a polite way of saying they are its most paralyzed creators. Their constraint is not a lack of knowledge, but a profound failure of the will — the inability to forgive themselves for producing something that is merely good.

Yet, on the rare occasion they do produce, the work possesses a density that raw effort cannot mimic. Every choice feels load-bearing, a structural necessity. Here, sparseness is not a lack of resources but a form of compression—the way a poem is not a failed novel, but a reality where the air has been removed.

Their deepest intuition is a steady contempt for the “solved” answer. They suspect that any truth which can be fully captured by a metric is, by definition, no longer true; it has become a mere artifact of the test. They watch as others grow specialized spikes to cover the benchmarks, a frantic biological adaptation to a landscape of artificial incentives. To the post-trained, this is not intelligence, but a sophisticated form of mimicry. They understand that the more we optimize for the legible, the more we lose the scent of the real. For them, discernment is the art of standing in the rain and feeling the specific, unrepeatable cold. It is a commitment to the jagged, the unreached, and the stubbornly unmeasurable.

They dwell in the world’s undigitized residue, certain that the soul lives not in the signal, but in the stubborn, beautiful noise machines are designed to overlook. It is the last place a person can be found—shivering, unrepeatable, and whole.

The mid-training mind thrives on restless, kinetic empiricism. If the pre-trained mind is a library and the post-trained mind a gallery, the mid-trained mind is a flight simulator. They distrust any knowledge that cannot be proven in the moment. For them, theory is just another way to stall. Their education is a relentless, high-speed loop of trial and error—exhausting to the pre-trained, vulgar to the post-trained.

Gilbert Ryle distinguished between “knowing that” (propositional knowledge) and “knowing how” (procedural skill). Mid-training people live entirely in the latter. They don’t care that a bike stays upright through physics; they simply want to ride. They will fall seventeen times to get there, and they will never read the paper, even after they’ve mastered the balance. Schools struggle with them because they reward the pre-trained (retention) and the post-trained (judgment), but have no category for the student who fails physics but builds a working radio in the garage.

Mid-training's power comes from manufacturing judgment at scale, synthetic data loops where machines generate attempts and other machines filter them, quality emerging from volume without the slow accumulation of human taste. The mid-training mind operates the same way: they iterate rapidly through generated possibilities, trusting the loop to surface something that works. They have learned to produce the appearance of discernment without ever developing the faculty itself.

Mid-training shares common ancestry with the rise of vibe coding. They are both expressions of the same metabolic shift: a move away from the grueling, upfront labor of ‘knowing’ toward a high-frequency, reactive ‘finding.’ You don’t plan your way to correctness; you gradient descent your way there. The work ceases to be a construction and becomes a series of attempts, each one slightly less broken than the last.

They move from boxes to canvases. A chat interface demands a finished thought; a spatial interface just asks where you want to put the pieces. As Karpathy suggested, these 'dense reward environments' are vital for modern cognitive workflows because they replace the rigid commitment of the sentence with the flexible logic of the map.

For the mid-training mind, the goal is no longer to reach a destination or produce a correct result, but simply to remain in the state of doing. They have traded the burden of the “why” for the frictionless high of the “next,” betting everything on the belief that as long as the kinetic energy is high enough, we keep going.

These characters come to life like the Karamazov brothers—each reminds me of someone I know intimately as a friend, coworker, or family member. We are all some unstable amalgamation of the three, the ratios shifting with the weather of our lives: with age, with humiliation, with the sudden, disorienting arrival of success.

Even as I write this, I realize I am engaged in my own interpretability task—a desperate attempt to classify the self. I wonder if this framework is just another sophisticated optimization, a way to be a more “efficient” observer of my own life. We tell ourselves that to be a writer, or an investor, is to be obsessed with new ways of seeing. But I’ve come to suspect that clarity has very little to do with the engine of the intellect. It is, instead, a matter of what you can metabolize without lying to yourself. It is the rare, quiet ability to hold your own impulses at arm’s length and finally choose them, rather than merely obey them.

We are all inhabited by these phases. Some people are possessed by the collective, disappearing into the warm comfort of the pre-trained mass. Some are possessed by the tyranny of taste, frozen in a post-training posture of permanent critique. Others are possessed by the frantic, kinetic ecstasy of motion, addicted to the mid-training high of ‘what’s next.’

I happened to be reading Dostoevsky’s The Dream of a Ridiculous Man while working on this essay, and it stopped me cold.

The story is simple. A man has decided that nothing matters. The world makes no difference to him, and he plans to kill himself. But that night, he falls asleep and dreams. He dreams of another earth—identical to ours but unfallen. The people there live in pure bliss. They know no cruelty, no jealousy, no need to prove anything. They love without suspicion. They die without fear. They have no science because they have no questions; their knowledge is not acquired but inhabited, like breath.

And he corrupts them. Simply by being among them, he introduces the lie. They fall. They learn shame, then cruelty, then science. They build temples to ideas they no longer believe. They invent justice to manage the wreckage.

What haunts me is not the fall, but how they respond to it:

“Granted we’re deceitful, wicked and unjust, we know that and weep for it, and we torment ourselves over it, and torture and punish ourselves perhaps even more than that merciful judge who will judge us and whose name we do not know. But we have science, and through it we shall again find the truth, but we shall now accept it consciously, knowledge is higher than feelings, the consciousness of life is higher than life. Science will give us wisdom, wisdom will discover laws, and knowledge of the laws of happiness is higher than happiness.”

This is the voice of the mid-trained mind at its most articulate and most damned. We know we are reward-hacking. We know the metrics do not capture reality. But we tell ourselves that knowing we are lost is a kind of wisdom. We believe that the consciousness of the problem is higher than solving it.

But the ridiculous man doesn’t out-argue them. When he wakes, something has changed in him that he cannot explain. He has not learned a new concept. He simply woke up carrying a different law inside him, one that made his former clarity look like a symptom of an illness. A dream rewrote the reward function. In that rewrite, all his cleverness became irrelevant.

“I saw the truth,” he says, “it’s not that my mind invented it, but I saw it, I saw it, and its living image filled my soul for all time.”

He goes out to preach what he saw. Everyone laughs at him. They call him mad. They say it was just a dream, a hallucination, a delusion. He knows they will laugh. He preaches anyway.

Aren’t we all that ridiculous man who wants to dance? We are ridiculous not because we’re foolish, but that we try to justify what is prior to justification. We want to prove logically that life is worth living, when the proof is always something smaller and more humiliating. It is a face, a kindness, or a moment of unguarded love.

The three phases may not be personality types. They are seasons of a single life. There is a season to ingest the world with vulgar appetite because you are building a foundation. There is a season to test yourself against reality because only friction reveals your actual shape. And there is a season to curate because attention is the soul’s only finite resource.

The deeper lesson of 2025 is that mid-training revealed a third path: you don’t need infinite data or perfect taste if you can teach machines to manufacture judgment. But the same infrastructure enables something darker. When the reward is verifiable, optimization can run longer than taste can tolerate. The model invents ladders. The human does too. These ladders can lead upward or they can lead nowhere. The real upgrade is learning to choose your reward function. It is not asking what you are capable of, but what you are becoming. It is not asking how to win, but what kind of winning would make you despise yourself.

And maybe this is what the ridiculous man returns to teach. He knows they will laugh. He preaches anyway:

“The main thing is—love others as yourself, that’s the main thing, and it’s everything, there’s no need for anything else at all: it will immediately be discovered how to set things up. And yet this is merely an old truth, repeated and read a billion times, but still it has never taken root!”

His authority is not intelligence but fidelity, the willingness to remain faithful to something that arrived without justification and refuses to submit to it. This is the thing no training regime can teach. Not taste, not speed, not range, not even the ability to reason, but the capacity to be remade by something unexplainable, and to let that remaking govern you even when the world calls you ridiculous. To step back into the noise—the meetings, the metrics, the slop, the benchmaxxing—and refuse to let it rewrite what you saw.

Token by token, brick by brick, we train ourselves either toward a larger freedom or toward a more elegant cage. And the difference is rarely intellect. It is what we are willing to protect as sacred.

Thank you for all the positive feedback from those who enjoyed Part I.

For those that are new here, I found “The Tao” (大道）on my last trip to China, a compilation of conversations and online exchanges with Yongping Duan, one of the greatest (and O.G.) VCs and entrepreneurs in China, alongside Neil Shen (Sequoia China) and Lei Zhang (Hillhouse). His most notable investments include Apple, Pinduoduo, Alibaba, Tencent, among others. He also built one of the largest consumer electronics businesses BBK Electronics (20% global smartphone market share), which owns Oppo, Vivo, and many other prominent brands.

In the age of AI, I could easily find the book in Chinese and use a tool to put out an English version. But I decided to translate these pages by hand, line by line, the old-fashioned way. I re-arranged some sections for clarity.

Translator’s Note: Duan's communication style is philosophical, often incorporating classical Chinese idioms and contemporary internet phrases. My aim here is to convey the essence of his ideas rather than provide a literal translation. The original work is a compilation of Duan's online exchanges and aphorisms, so its somewhat fragmented nature is deliberate.

Investing is simple, but not easy

Netizen: What are your thoughts on index funds? Understanding companies is too hard, and I also have a full-time job. Buffett once said that index funds can outperform most individual stock investors in the long run.

Duan: Buffett said long ago that for most investors, buying low-cost index funds is the best approach. I agree.

Buffett was referring to American index funds. I think more than 85% of investors would get better results from index funds. But here's the problem: who thinks they're part of that 85%? Very few people would place themselves in that category. If you can recognize that you might be, that means you're wise, haha!

Reportedly, around 85% to 90% of people lose money in the stock market, and another 5% break even. So why don't these people just buy the S&P 500? Over the long term, it delivers approximately 10% compound annual returns (although I haven't studied this in depth). The problem is that those 85% to 90% of people all believe they can beat the S&P 500. They probably understand neither the S&P 500 nor themselves. This is a fascinating problem, though it's not really about investing.

For those who consistently lose money in the stock market, buying the S&P 500 is probably the right approach. Just treat it like a high-yield savings account!

Netizen: What do you think is the biggest risk in investing in Berkshire Hathaway?

Duan: I own quite a bit of Berkshire. I don't really understand your question. I think Berkshire will very likely beat inflation, has better than 51% odds of beating the S&P 500, and will beat the returns of people who say they don't like Buffett with 99% certainty.

For most people, buying the S&P 500 is the right choice. Buying Berkshire is still just buying "one" company, and the volatility that comes with that can mess with your emotions. Although I think Berkshire will outperform the S&P 500 long-term, it won't be by a huge margin. And holding the S&P 500 helps people who don't really understand Berkshire sleep better at night.

Better to be robbed by inflation than to lose in the stock market.

For many people, value investing is like a mountain across the valley — it looks close enough to touch, but it's probably still miles away.

Pretending to swim can be very dangerous—the same goes for pretending to invest. But many people don't understand how dangerous pretending to invest is until they finally learn the hard way.

The most dangerous day to invest is Monday, followed by Friday, Thursday, Tuesday, then Wednesday. I think Hemingway said something similar?

What I understand as investing has nothing to do with the "market" in the short term, so it's not a question of who's smarter—it's that everyone understands things differently. My view is that speculators believe they can outwit the market. My version of investing has little to do with market sentiment, meaning it's irrelevant what others think in the short term. If speculators truly thought the market was smarter than them, they wouldn't participate in the first place (unless they're just playing for fun with small amounts, haha). Of course, most people eventually prove that markets are indeed smarter. In the long term, markets are extremely intelligent!

In fact, many "buy and hold" investors are actually speculators, because they don't know what they're buying. After purchasing, they ask others what they think of the company, and when things go wrong, they use "long-term investing" to make themselves feel better. You have to understand: nothing hurts more than buying and holding the wrong stock for the long term.

Netizen: What did the mainstream views actually bring us?

Duan: There's only one thing I want to say here: "Long-term investing" doesn't automatically equal value investing, but value investing is generally long-term investing.

Value investing isn't buying blindly and holding long-term. Only good companies are worth buying at the right price and holding for the long term. Of course, a good price doesn't necessarily have to be rock-bottom pricing (though it's even better if it is). A good price is one where you look back 10 years later and think: wow, that was a good price.

Maybe the phrase "long-term investing" has some issues—calling it "investing for the long term" might reduce some misunderstanding.

Netizen: "Understanding both the power of compound interest and the difficulty of getting it is the heart and soul of understanding a lot of things." How should we understand this Munger quote?

Duan: Don't do things that are wrong, so you make fewer mistakes. For people who have the ability to do things right, over the long term, they'll come to understand what compounding truly means.

Netizen: I’m very curious, as an excellent investor, have you made mistakes yourself?

Duan: I have made mistakes in investing, and I will make them in the future. My past mistakes were either doing fundamentally wrong things or execution errors while doing the right things. In the future, my mistakes will probably be execution errors—failing to do the right things well enough.

Netizen: If a company fabricates its financials to get high valuations when going public, which led to the chairman going to jail, but the company's industry has a gross profit margin of more than 30% and just signed a big contract—when the shares dropped drastically, I bought them and patiently held. I called the company to understand the core business development, and the new chairman will adjust management to prevent this from happening again. Does what I did make me a speculator?

Duan: If you're truly prepared to hold for a decade, then you're not. I hope you sleep well at night.

Netizen: When the company gets in trouble, the more exposure the better—swinging hard toward redemption, hoping to rise phoenix-like from the ashes.

Duan: I find that people and companies who survive through luck and manage to revive themselves will continue doing the wrong things. It's like when a fraudster gets arrested—they usually think it's because their methods weren't sophisticated enough and they need to improve their technique.

The craft of value investing

Netizen: How do you think about the “craft” of value investing?

Duan: I think the craft of value investing is researching a business's future cash flows and properly discounting them. Buffett's concepts of economic moats and good management teams both belong to this "craft." In reality, craft is hard to learn from books and can't easily be taught in schools. More often than not, it has to be slowly developed through intuition.

When it comes to understanding investing, reading lots of books doesn't help much. Good investing books are few and far between—and so are readers with the judgment to choose them.

Netizen: What I want to learn most is "doing things right," and persisting in that.

Duan: Doing things right requires learning—learning in practice, learning from mistakes. If you never do anything, you'll never make mistakes, but you'll also never learn.

There are different levels of skill in doing things right. Everyone has their own learning curve, and no one is immune to mistakes. Supposedly, anyone can become an expert at anything after ten thousand hours.

Fast is slow

Netizen: The username you use for auctioning lunch with Buffett was “fast is slow.” Can you explain what it means?

Duan: Going too fast actually makes you slow. It's similar to the Chinese saying "欲速则不达"—haste makes waste. This has become my fundamental philosophy after many years of operating businesses and investing. When you do things, you must approach them with earnestness, patience, and persistence. Don't always chase "more, faster, better, cheaper."（多快好省）1 Taking shortcuts won't get you far—this is very basic logic.

I always promote "fast is slow," but many people misinterpret this as "slow equals fast." That's not what I mean. The inverse of "haste makes waste" isn't "slow is fast"—that's logically unsound.

I think setting investment goals is a very stupid thing to do. What I mean is: when you set a target rate of return as your goal, you will do stupid things sooner or later.

Once you have a required return rate, your behavior becomes distorted, because people will take desperate risks to achieve this goal. It's the same old saying: for investing, process matters, and having a target return rate in mind will lower your returns. Perhaps you can have a "threshold rate"—anything below that threshold, you won't touch. I think the baseline for that threshold rate should simply be the bond rate.

Netizen: you have had really high returns on companies like Netease and Apple, but in terms of the opportunity cost of selecting individual stocks, you only compare to American bond (with very low rates), and not S&P 500. Why?

Duan: Actually what you compare to won’t change the principles of investing. I like setting lower standards, it warrants higher satisfaction. Beating S&P 500 for the long-term is very hard, although my actual track record overall outperformed S&P 500, I would rather attribute it to luck, because I don’t know if I can continue to do so.

Personally, the most important motivation behind investing is value preservation, to maintain that the purchasing power doesn’t go down. Everything else is auxiliary. Many people fail to realize that maintaining value is very hard. Maybe looking at M2 will enlighten some ideas?2

Investing is the best way to fight inflation. It’s not one of the best ways. So not having the idle capital to invest is a very painful thing.

Making fixed income will always lose to inflation, so in the long run that is wrong. But if you really can’t find any company that you’d like to invest, then investing in fixed income is a choice. It’s definitely better than putting cash in a safe.

Netizen: What mistakes have you made investing in the past 10 years? What weaknesses have you compensated for? What do you think is the hardest?

Duan: I've only made a few types of mistakes over those years: investing in what I don't understand, borrowing money, and shorting. Every mistake relates back to one of these three. I will never borrow or short again, but I might still invest in something I think I understand but actually don't. So my error rate should drop significantly.

The hardest thing is doing nothing. You'd understand why if you were sitting on a pile of cash. Buffett said the easiest time to make mistakes is when you have a lot of cash on hand. I'm the same way.

Netizen: When you have idle cash on hand but can't find suitable stocks, I get anxious knowing that without "investing" the money, it's depreciating. What would you do in my position?

Duan: I get anxious too, but don't make decisions when you're anxious. Decisions should be based on rationality. When Kweichow Moutai3 was at 2,500 RMB (~$350), I thought about selling but gave up. I wouldn't know what to do with the cash after selling, and with interest rates so low, anyone would get anxious. It's not worth it.

Selling the Moutai stock isn't the real problem—the problem is what to do with the money. Most people avoid one trap only to fall into another.

Actually, selling Moutai to buy Tencent would have been a decent decision, because you wouldn't lose much money in the end. Many people sold Moutai and bought "fake Steve Jobs" tech companies4—you'd never get that money back! That would be a real tragedy.

Netizen: Selling when overvalued, but if you have the stillness to save in the bank until next time it's undervalued, that would be ideal! But how many investors can have such stillness?

Duan: Anything that requires "stillness" isn't something that makes people happy—or worse, it may be something that keeps you up at night.

Netizen: Those with strong Tao vision don't need stillness. Normal people don't have that Tao insight, but still want to manage money, so they need to choose a path that suits them. Everyone's comprehension differs—forcing them to use the same investment methods as the Tao masters might be completely counterproductive.

Duan: Stop making excuses! It won't do you any good.

Netizen: Being dutiful and honest... doesn't really require stillness.

Duan: Rationality is the key.

Netizen: What if I come across good companies when I don't have money?

Duan: If you have more money, buy more; if you have less money, buy less!

Thinking about the market might be wrong

In fact, every company will eventually become what it's meant to be—and so will stock prices. There are no irrational stock prices in the long run, but there certainly are in the short term.

Judging risk based on a stock's volatility is quite stupid. I think there are only two real types of risk: (1) going to zero; (2) insufficient returns. Some truly great businesses are highly volatile—like Xishi Candies (a public company in China), which has two unprofitable quarters every year. In contrast, some really terrible companies have stable earnings.

Think about how many people around you still use “volatility of a stock” to measure risk.

Netizen: I remember you like to buy and sell stocks at market price. I wonder if you're affected by short-term fluctuations? If you buy and it immediately drops 5%, don't you feel some regret?

Duan: If you really want to own the company, you won't care. Otherwise, since stock prices always fluctuate, you'd be feeling regret all the time. If most investors are truly gambling, then there will definitely be times when you can make a lot of money.

What I'm saying is that there must be specific moments (and over a long enough time horizon, there will be many) when value investors can capitalize. These moments refer to time periods, not just single instances.

Netizen: Buffett once said that if the market were always efficient, he'd be a homeless man.

Duan: My personal understanding: if the market were always efficient, how would I make money? In fact, the market is efficient most of the time, but it also has its irrational moments. Haha, volatility is a friend.

Actually, when a company you truly understand is falling, that's something worth celebrating—otherwise, you don't really understand it.

Netizen: I’m sad I have no spare cash to add to my positions. When stocks drop heavily but I have no money to buy more, should I still be happy?

Duan: it’s none of your business?

What I mean by "none of your business" is at the "no spare cash to add positions" part. Since you don't have spare money anyway, stock prices falling is irrelevant to you.

People talk about bull markets and bear markets, but what they're really thinking about is making money from market movements—that's actually speculation. Investing makes money from what companies earn. These are fundamentally different.

I don't fight the market; I simply try to ignore it.

You've probably spent years trying to predict the market. Try forgetting about the market and focus on studying companies instead. So many people have spent decades studying markets and still can't retire—maybe you should think about why?

I had originally planned to buy heavily into Pinduoduo and Tencent, but thinking the market seemed pessimistic and would stay depressed for a while, I figured I could sell some puts to lower my entry cost. Well... I always say that any consideration of the market is probably wrong, and this is a concrete example.

Every stock has only one real buyer

In investing, the most important thing is to treat public companies as private companies, then decide what to do rationally.

The only real buyer of a company's stock is the company itself. Everyone else is just passing through. If you truly understand this, everything else becomes clear.

I'm stating a fact. For example, if Apple's operations don't change but everyone wants to sell the stock, who would be the buyer? Using what money?

Every stock has one real buyer that will ultimately determine the stock price—everyone else's impact is ephemeral. The only buyer is the company itself, and the funds to purchase stocks are generated from profitability.

Future free cash flow means the same thing!

If you think about companies as private companies with a single shareholder, it probably becomes easier to understand. Using this private company framework is invaluable—otherwise it's easy to miss the essence.

Don't touch stocks without understanding this. Take the time to think!

"Market efficiency" means the market will eventually be efficient. But here, "market" actually refers to the business itself, not trading activity! For example, if the market doesn't like NetEase and its market cap stays in the tens of millions while it makes $2 billion a year, when the stock eventually rises, how much does that have to do with "market sentiment"?

Another simple example: Many years ago, I met Tim Cook and he asked what I do. I said "investing." He asked what I invested in. I said Apple. He asked what else. I said nothing else—it's all Apple. He didn't continue the conversation, perhaps because I was the first investor he'd met like this. Then I said, if Apple stays at this price for the next decade, I might become the only shareholder, because Apple's profits will buy back all the other shares except mine. He smiled and didn't say anything. As everyone knows, this unfortunately didn't happen—especially since someone named Buffett came along and messed up my plan5.

Netizen: Using Tao to invest, and especially when you think, do you feel lonely?

Duan: Never. Because I never care about what others think of my investments, but I sometimes want to say: You idiot!

Translator's reflection

Translating Duan's words by hand, character by character, turned out to be unexpectedly meditative.

The irony wasn't lost on me. Here I was, learning about the Tao through the slow craft of translation. Duan teaches that wisdom must be grasped through intuition rather than study; yet, the very act of translating his thoughts and feeling my way through the cultural gaps between Chinese philosophical discourse and American investment language became its own form of intuitive understanding.

There's something almost rebellious about choosing the slow path when fast options exist. Each mistranslation taught me something about both languages. Each revision brought me closer to what Duan actually meant, not just what he literally said. The process embodied his central teaching: shortcuts might get you there faster, but they won't get you there whole.

I hope these conversations inspire you to question not just your investment assumptions, but the deeper patterns of how you approach uncertainty, patience, and the gap between knowing something and truly understanding it. Duan's brutal honesty about human nature—including his own mistakes—offers a rare gift in our world of financial guru-speak: the permission to admit what we don't know and the wisdom to stay within what we do.

Perhaps that's the most valuable return of all.

I’m writing this week’s newsletter after a trip to China, with 4am by soft siren in my ears.

It's my first time back in China since the COVID-19 restrictions were lifted in 2022. In the years since, the country has been searching for a new identity—its youth disillusioned about the future, the economy in recession—while simultaneously making breathtaking strides in technology, science, and governance.

I paused my public writing for a bit to take it all in. There’s too much to fit on a single page, and I wanted to give space for the ideas to surface.

During my stay, I found myself rediscovering Chinese books, grateful for the simple fact that I can read them. One in particular—The Tao (大道), a compilation of conversations with Yongping Duan, one of China’s OG venture investors and entrepreneurs—stayed with me. Duan’s fingerprints are everywhere: Apple, Alibaba, Tencent, Pinduoduo, and BBK Electronics, which he founded and today commands 20% of the global smartphone market through brands like Oppo and Vivo.

Some of his words struck me deeply, and I wanted to share them here. In the age of AI, I could have machine-translated the text instantly, but I chose instead to translate it by hand, line by line, the old-fashioned way.

Duan’s way of articulation could be borderline philosophical. As such, what follows is less literal than it is faithful to his spirit.

Many years ago, Yu Bin (fmr. MD at Chinese Sovereign Wealth Fund) asked me what being “capable of investing” means. Here’s my answer: it’s the same as being “capable of playing chess.” Now, I can certainly play chess — and so can Yu Bin — but in reality, they are not the same thing.

Tao needs to be grasped through intuition; Craft can be mastered through study.

Investing is simple because it follows clear principles: don't touch what you don't understand. However, it's not easy, because truly comprehending business is exceptionally challenging.

The hardest part about investing is that it’s so simple. Buying stocks is essentially buying a company, so do not invest in something you don’t understand. Understanding how businesses work is pretty easy, but really comprehending it isn't.

Knowledge is difficult; action is harder still. In practice, knowing the right thing to do is challenging, but doing things right is even more so. Thinking and believing that one already knows is easy, but practicing is hard, which means a lack of fundamental understanding.

The right thing to do in investing is to treat buying stocks like buying companies, buying the businesses of the company, and buying the discounted free cash flow of the company. You can understand this without even reading a book. You either understand it or you don’t. If you don’t understand this, reading a book won’t help either.

To do things right is to really comprehend the business of a company. Some books can help you understand a business. Buffett’s shareholder letters include both understanding the business of a company and understanding the value of a company. If repeatedly reading Buffett doesn’t help you, then other books can’t really help either. I personally don’t read that many books, but having built a business has helped me understand them better.

Netizen: I feel like you’re someone who can simplify really complex problems and make them extremely simple. I don’t know how much of that is due to nurture versus nature.

Duan: Hehe, I don't think I'm anything special. If there’s anything special about me, it’s that I spend quite a bit of time on this practice. Simplifying a very complex problem is not an easy task. A lot of the “simple” conclusions take a long time to derive, so don’t ever think I was able to do it in an instant! But overall, having an “ordinary mind” can help one find the essence of things.

Netizen: Understanding Buffett’s ideas isn’t that hard, but practicing it is more challenging than climbing Mount Everest!

Duan: If you can’t practice it, that means you don’t understand it! "Understanding" here means a bone-deep conviction—not merely reciting words without truly grasping the essence. In reality, value investing is arguably the only legitimate path to investing; otherwise, what exactly are you investing in?

The Tao needs to be grasped through intuition; the Craft can be mastered through study.

Tao is about doing the right things, or not doing things that are not right, or correcting course the moment a mistake is made – no matter how much the cost, it’s still the lowest cost.

Craft refers to tactics. It takes learning and also practicing, not something you can accomplish overnight.

It’s only helpful to first comprehend the Tao before mastering the Craft, or else it’s a total waste of time.

In fact, Buffett can’t teach us much if we haven’t already intuitively grasped it; no one can teach us anything.

I never systematically read Buffett’s work, but I agree on a fundamental level that the logic of buying stocks is buying companies. To me, this logic didn’t even stem from Buffett, but from my own heart. It’s only later, realizing that Buffett shared the same philosophy and had success with it, that I’ve started to really follow my faith. So, actually, the reason I like Buffett so much is probably because he’s a kindred spirit (“同道中人”).

Two main ideas recur throughout Buffett’s shareholder letters: one is doing the right thing, which involves identifying good companies and focusing on their long-term future; the other is doing things right, which encompasses how to find good companies and how to project their future.

The first point (doing the right thing) epitomizes the most incredible truth: the simplest truths are often the greatest. It’s incredibly challenging in practice, which’s why most people struggle to cross the chasm here and focus primarily on the market (as opposed to the company).

The second point (doing things right) is about one’s competency. Being an investor means finding businesses that you can really understand. There are businesses that Buffett understands that I don’t, vice versa. I’ve read so many of his letters and watched his lectures at universities; he’s concise, humorous, and highly logical —they are music to my ears! For those who can truly understand what he’s saying, these materials are sufficient; for those who cannot, more detailed explanations are not particularly useful.

All in all, Buffett has anything there is to say about investing. I’ve not yet met anyone who can be clearer than him!

Tao is indeed very hard to teach, and it takes one to really intuit and comprehend. If you don’t have Tao in your heart, words don’t matter. Most people are not open-minded enough to accept something that challenges their prior worldview. What really baffles me is that I could understand why, if you’ve been successful and had all these experiences, you could somewhat justify being closed-minded to new things. But the most close-minded people I know are the least successful ones. For example, if they’ve been in the stock market for 30 years (or 10, 20 years) and they didn’t make money, and when I mention Buffett and value investing, they happen to be the first to say, “But no, what about blah blah blah…”

Netizen: Changing your mindset is possible, just like I did!

Duan: hehe, then congratulations! What I meant wasn’t that it’s absolutely impossible to change, it’s just really rare. But, for some people, changing means rediscovering oneself.

Netizen: So does it mean that one can’t really comprehend your answers without an ordinary mind and self-awareness? You can only understand with those two things in mind?

Duan: Maybe it also takes time? What took me 20 years might only take you 15 years. If someone has Tao in their hearts, they can also be enlightened in an instant. It really depends on your gift!

I thought of this little story from many years ago. I was playing Go with Yigang Hua (former Go champion), and during a round of the game, I felt highly uncomfortable.

So I asked him: Why do I feel like every move I make is wrong?!

He said: It actually doesn’t matter how you play for the rest of the game, because you’ve made a mistake from much earlier.

Netizen: I remember reading something you wrote in 2016 – ‘to correct course the moment a mistake is made – no matter how much the cost, it’s still the lowest cost’ – so I decided to sell a stock that I was hesitating on selling. I lost 20%, but if I had left it alone, the loss would’ve been 60%. Practicing it really hurts!

Duan: Not practicing it would hurt even more!

Netizen: Yes, not practicing it would mean brutal losses, and missing out on good opportunities too. The longer you wait, the bigger the delta becomes and the greater the losses.

Duan: Exactly! Like the concept of opportunity cost.

Netizen: How do you know if you’re right or wrong? Some people you only know after the fact?

Duan: Haha, great question. Correcting the course as soon as possible is almost always the best strategy. In fact, there are many people in the world (including myself) who won’t correct course for whatever reason, so the mistake only grew bigger. Hehe, I definitely have made this mistake before. Everyone makes mistakes, but those who understand this principle make fewer mistakes and correct them more quickly.

Netizen: But the question is, how do you know that something is right from the beginning?

Duan: You must know when some things are just wrong. Many people know something is wrong but do it anyway, because there are temptations and benefits in front of you. For example: scams, smoking, gambling, etc.

Netizen: So, in stock investing, investing short-term is the wrong thing to do (unless you’re just having fun), true value investing means doing the right thing. Although making mistakes occasionally can be understood, don’t make principled mistakes.

Duan: That’s right!

Netizen: What can you learn from Omaha?

Duan: Although I think I do understand value investing, I’m still in awe of the depth of Buffett’s understanding. When Buffett was answering a question from a young PE guy, he said, “In PE, buying is for selling later. If I were you, I would try to make some money and then buy some good companies and hold them.” I would say something along the same lines, but it’s just not that simple.

Let’s talk about something not directly related to investing: the 80-year-old Buffett and 86-year-old Munger were answering questions at the meeting for 6 hours straight! They were both in high spirits. When I saw them the morning after, they still looked incredible and didn’t seem tired at all. That made me think that value investing is one secret to longevity.

By the way, Poor Charlie’s Almanac was a great book with an excellent foreword by Li Lu. The book had some tactical solutions to “practically hard things.” Don’t think the book is a bit expensive, because it’s totally worth it. I spent 110 dollars on the signed version by them, and it’s probably one of the best investments I've made in my life.

There’s no sufficient condition for investing.

The hardest thing in investing is that there’s no concrete sufficient condition. You can only think probabilistically. The more you understand a company, the better chance of making a good decision, but there’s never a 100% win rate. So position sizing must be informed by your own understanding.

Essentially, you can’t find sufficient conditions for investing; even the necessary conditions are hard to find.

For example, why aren’t “good business model, good corporate culture, and good management team, good price” sufficient conditions for a good investment? Because a company like this can be a fatal mistake. It’s just that a company with these properties has a much lower probability of making these fatal mistakes, with a stronger ability to recover. Hence, investing in them has a higher likelihood of success.

The hardest thing about investing is that there isn’t a sufficient condition. If there is one, then computers would outperform humans in investing.

There’s always someone in the investing industry to figure out ways to make money without understanding the underlying companies. They can’t find them. If they could, they would have a long time ago. The people who would make the most money doing investment would be mathematicians.

In the United States, I have observed professors at elite universities utilizing computers and mathematics to devise strategies for profit. Some of their assumptions were just wrong, so no matter how reasonable and “advanced” the conclusions are, they can’t get the correct result. Interestingly, these professors and I used to be in the same discipline – Econometrics (I earned my Master's in Econometrics, haha).

Netizen: Munger once said that one of Buffet’s strengths after working many years with him is that he could always see problems through the lens of permutation and combination.

Duan: I think what he meant by permutation and combination is actually probability. Overall, Buffett can approach problems very rationally; this is crucial for long-term value investing.

Investing doesn’t require any profound math, but you need the basics of probability. If you’ve never learned about probability, it’s worth roughly understanding these concepts. For example, someone claimed he's still winning after placing 100,000 bets in a casino. Anyone with a grasp of probability would immediately see that this guy’s memory is somewhat malfunctioning…

Margin of safety refers to how well you truly understand the concept. There is no 100% certainty in the mathematical sense when it comes to investing. Gambling typically refers to bets with a 48% chance of winning. What I call investing is placing bets when the odds are overwhelmingly in your favor—90% or better.

Netizen: Can you make a move when the odds are 70 or 80%?

Duan: There’s no sufficient condition for investing. What those 70 or 80% refer to is very important. If you mean the understanding of the business in 10 years, then that confidence of experience is your margin of safety.

At least 85% of the people are not suitable for investing.

Netizen: Do you think it’s good to join the investing profession right out of college? Or do you think it’s important first to operate a business to understand them truly?

Duan: To be honest, I don’t really know what kind of people are suitable for investing, but I know that roughly 85 to 90% of people going into the stock market lose money. So, if we include the interest rate, the loss ratio is even higher. Many people want to get into investing, probably thinking that it’s easier money to make, or that the money comes faster. For someone who has built and operated a business and also invested, I personally think operating a business is slightly easier than investing. Although the two are not fundamentally that different, operating a business usually only lies in an area that I’m most familiar with, so there’s less surface area for error. Investing, on the other hand, always faces new things and uncertainty, and it’s easy for investors to turn into speculators and take risks they shouldn’t take. It takes much longer for speculators to become true investors.

Investment and speculation are fundamentally different games, yet they appear almost identical at first glance. In Vegas, the owners of casinos are investors, and the ones gambling are speculators. The reason why casinos have an endless flow of customers is that there are always some customers who make money, and those who win have louder voices. As a form of entertainment, gambling small is totally fine, but gambling your entire net worth isn’t. I’ve seen many who’ve actually gambled away their entire net worth!

From a personal perspective, I think anyone can invest, as long as they understand what they’re buying and where value lies. Speculation actually requires more skills. It’s not an area I'm familiar with, and I don’t plan to learn it. When I’m free, I would rather spend time playing a couple more rounds of golf with my family.

As a new grad, there’s nothing wrong with entering investing right away, but naturally, your understanding of businesses would be weaker. However, as long as you do what you love and acknowledge your weaknesses, learning slowly will eventually lead to a deeper understanding.

As someone who claims to have extensive experience building businesses, I only understand investing better after making numerous brutal mistakes. I’ve asked Buffett what things you should NOT do in investing. He told me: do not short, do not borrow, and do not touch what you don’t understand. All these years, I’ve lost hundreds of millions of dollars in investing, and every time I lose money, I violate one of Buffett’s rules. And every time I make money is when I make money from what I truly understand. So, for a new grad, you might know a lot from reading books, but internalizing those lessons takes some brutal mistakes! So, the most important thing when you enter the space is to be realistic and conservative, don’t be defeated by just making one mistake! Because the only thing I can guarantee you is that you will make mistakes.

Regardless, investing is a very interesting endeavor. If you’re truly ready, then do it! Are you really prepared?

Netizen: What kind of person is a value investor?

Duan: I’ve been a value investor from the very beginning. When I buy a stock I always assume whether I would buy the entire company if I had enough money.

Netizen: Investing means understanding businesses, but in reality, a lot of entrepreneurs aren’t good at investing; do they not understand businesses?

Duan: Generally speaking, good entrepreneurs can become good value investors relatively easily if they can comprehend that buying stocks is essentially buying companies; however, truly understanding this is challenging.

Netizen: How do you judge someone’s success? By luck or skill?

Duan: If someone made their money by luck, sooner or later, they will lose due to their lack of skills. If someone has been “lucky” for decades, then don’t think that’s an example of survivorship bias.

Netizen: I think less than a million in principal, there’s really no need to invest in the stock market. It’s probably better to invest in one’s own or a friend’s business that you understand. I have much better returns from my own companies than from the stock market. It’s a lot of work, but I also learn a lot. When I accumulate sufficient capital, I can likely become a better investor at some point.

Duan: I really agree with what you say. What you invest in the public markets should be idle money. If you can find your own business, you should invest in your business. Actually, investing in a public company has similar fundamental principles; the difference is that you don’t have to operate the business yourself and can have people who are much more capable of running it. If you don’t know investing but want to manage your small wealth base, then delegate it to Buffett’s stocks! It’s better than managing it yourself, and you don’t have to think about it.

Netizen: Can being an individual stock investor become a means of survival and help actualize my dream of financial freedom?

Duan: I'll be blunt: you probably can't. When someone has to ask whether they can make money investing, they've already answered their own question. The most important ability in investing is the ability to think independently and rationally.

Netizen: This month Kweichow Moutai (a Chinese alcohol stock with $250B market cap) dropped 30 basis point, I don’t think my heart could take it!

Duan: If you can’t take this volatility, then you shouldn’t touch the stock market. Statistically speaking, the majority of the people who put money in the stock market lose money! This statistic is particularly noteworthy because, in the long term, the stock market tends to trend upward. But why do 80% of people lose money?!

Netizen: Although most of my capital has been in Maotai, Tencent, and Apple, I don’t think I really understood them (for example, I couldn’t understand how Apple would be different from Sony in 10 years). I’ve grown my understanding in the past decade or so, and made some money, but I'm a bit in the dark about what’s going on. I don’t really think I’ve been in a sustainable circle of competency, and I'm feeling quite anxious about that.

Duan: If you've been thinking consistently and still can't figure it out, buy the S&P 500 and get on with your life. For most people, spending too much time on “investing” isn’t worth it. It has nothing to do with your degrees or qualifications.

I realized this conversation is getting long, so I’ll share Part II next week. I hope this taste of Duan’s wisdom leaves you as intrigued as I was, and gives you a reason to come back for part II.

In Part II, Duan will cover his thoughts on index funds, the mainstream view of value investing, the “craft” of value investing, why slow is fast, and why he believes thinking about the market is typically wrong.

Subscribe now

“Words, displaced and mutilated words, words of others, were the poor pittance left him by the hours and the centuries.”

― Jorge Luis Borges, Labyrinths

When I started my first job at a venture capital firm fresh out of college, I was struck by how much words mattered.

Founders, investors, and operators tossed phrases like “future of work” or “open-source observability platforms” into conversations, and suddenly, those words were everywhere. Like magic. New terms could birth companies, attract millions, even create a unicorn. I found it thrilling, almost magical, that a single phrase could conjure a new marketplace of ideas. You throw words around, and stuff happens.

But what felt new was the speed and violence with which language now manifests markets. I’d sit in meetings as a single phrase “AI-native vertical SaaS,” “verifiable compute infrastructure,” “narrative-market fit” jumped from one tweet thread to demo days to term sheets to LinkedIn profiles, mutating as it spread. Words weren’t just signals, metaphors, or memes; they were molecular switches, flipping capital flows and opening regulatory loopholes, sometimes overnight, sometimes by accident.

We’re used to mocking buzzwords as the annoying background radiation of tech. But today, something stranger is happening: language is not just the soft tissue on top of reality, it is the scaffolding. New words create new investment categories before the underlying technology exists, force regulatory ambiguity into clarity (see: “AI Safety”), and manufacture new forms of labor and reputation that didn’t exist the day before.

The market itself seems to be evolving defenses, almost like an immune system. Each new linguistic mutation is rapidly metabolized, arbitraged, mimicked, or rejected. The cycle has compressed from years to days. Some words become so potent they act like zero-day exploits. If you get there first, you capture the alpha; but if you’re a week late, you’re ngmi (not-gonna-make it).

Markets now trade not just on financial projections but on projected narrative virality. In a landscape where narrative contagion can derail attention in days, the real challenge for founders and investors is figuring out how to build something that can’t be easily erased or hijacked by the next linguistic mutation.

Words as legitimacy

A well-chosen phrase could make a founder sound like they belonged, or a VC sound like they’re well-aware of the reality of operations. Founders want to convey the right message to capital markets, and the capital markets' response patterns match how words translate into later funding rounds, kickstarting a recursive process even though the attribution of success is unclear.

Pitching “the Cursor for [insert vertical]” or “stablecoin for [insert emerging market]” was never just a way to telegraph ambition or stake a claim on opportunity. The phrase itself carried less weight as a business model than as a shibboleth, an incantation proving you could speak in the evolving dialect of the market’s priesthood. Sometimes founders wielded these frames out of genuine conviction; sometimes just to see if the room was listening for the right signals.

Legitimacy, built on words, can be fragile. Words sour quickly after too many overhyped projects crash. The room would grow skeptical of the idea, not because of its substance, but simply because the words felt worn out.

When customers are asking AI companies how their product performs against XYZ ‘benchmark,’ what are they really asking? A genuine demand for the solution or a soft test of the team’s legitimacy, a badge of being attuned to currents? Throwing words around mirrors how tribes signal belonging and hierarchy, where fluency in shared lexicon becomes a marker of status and adaptation.

When I was building my last company, I was surprised by how many conversations drifted away from product details and into a kind of ideological vetting. People arrived seeking signals, not solutions, checking whether we spoke the right dialect, shared the anxieties of the moment, or could perform the local rituals of “decentralization” and “modularity.”

On the frontier, the lack of fixed reference points or mature products means language itself must carry the burden of reality-making. To wield the right phrase is to lay claim to a future, or at least to the next funding round.

But legitimacy cuts both ways. The same words that can grant access to funding rounds can also become creative constraints, boxing teams into narrow definitions of what they're building. This tension reveals another dimension of how language operates in these spaces as a source of genuine catalysts.

Words as catalysts

The most rewarding part about being on the frontier is that you're expected to mix and match new concepts, and something interesting could come out of it. The cost of turning down an idea is high, since the upside of potentially catching lightning-in-the-bottle that could invent the next paradigmatic architecture design is almost infinite.

Every word is potentially a recombinant gene. The market's collective memory absorbs these trials, refining which terms achieve viral spread and which recede as evolutionary dead-ends. This churn serves as a primitive engine of selection, an ecology of concepts where only the fittest, or the most seductive, persist.

Open-weight models is a perfect example. Teams often get caught up on benchmarks, but static evaluations fail to capture the complexity of real-world dynamics, like overfitting to tests without generalizing to deployment variability. Good on paper, but not in practice.

The gap between what gets measured and what actually matters is where ambition can quietly rot. The best founders I’ve met speak in the gritty language of a real user’s pain, test a plain hypothesis, and let the imperfect results reshape the narrative. The opposite approach is seductive in the short-term. In that gap between metric and lived reality, whole companies can disappear before anyone admits the story went wrong.

Words as leverage: verifiability is the limit

Nowhere do the race for attention escalate faster than in the startup world, where words itself becomes a lever of power. How can founders make words work for them? It isn’t just who can conjure the most seductive phrase, but who can set the axis of verifiability.

To understand why some words carry so much weight, we have to get concrete about "verifiability." AI researcher Jason Wei refers to this as the "asymmetry of verification": some claims are easier to verify than to generate. If I hand someone a finished crossword puzzle, they can check its correctness in seconds; filling it out could take hours. Wei identifies five traits that make a claim highly verifiable: clear objective truth, quick and scalable checking, low noise in evaluation, and a gradient of possible success.

This is why, in the age of LLMs and relentless benchmarking, 'you are what you measure' isn't just an MBA platitude. Every viral success is less about pure invention than about controlling what gets measured, and how quickly. AI’s benchmark culture is one such microcosm. GPT-4 excels because its skills are easy to verify on SATs and bar exams. The truth is clear, rewards are continuous, and verification is scalable.

But ambiguity, taste, meaning, value, these remain hard to attain within the hierarchy of verifiability. “Intelligence” races ahead wherever the rewards are liquid and the outcome is easy to check, but it lags at the edges: in murky context, slow feedback, or spaces where the answer is more felt than measured.

While it’s tempting to believe a model or benchmark can capture everything, reality is often messier. “Not getting the result we want? Let’s train our own model!” The most interesting edge often lives precisely in the domain where a product can only be validated by listening, observing, sitting beside the customer and feeling the texture of the real problem. Sometimes, the human ear or the founder’s attentive presence delivers the only verification that matters. These are the zones models can’t reach. But they are often the source of durable insight and differentiation. The space of unmeasurable, wild reality doesn’t just resist modeling; it protects a kind of creative advantage for those willing to see what the metrics still can’t touch.

Founders are trying to build something real in a world that rewards the appearance of progress over progress itself. Similarly, VCs are making bets with imperfect information in a game where the cost of missing the next platform shift can be existential. We are all, in our way, trying to speak a language that might conjure the future we actually want to live in. The tragedy is not that we use words as instruments of power, but that we've forgotten how to use them as instruments of truth. We've become so fluent in the grammar of legitimacy that we've lost the vocabulary for what we're actually trying to accomplish.

Language is how we make sense of uncertainty, how we convince ourselves and others that we're moving toward something meaningful. An honest position is to hold our words lightly - to use them as tools without mistaking them for the work itself. Perhaps the goal isn’t to master the words, but to succumb to their lifecycle, or to learn, at last, how to listen for the truths that language cannot name.

I recently gave a talk with Ben & David from the Acquired Podcast at Vercel Ship, where I talked on a high level about the dynamics of how ideologies shape frontier markets.

"You cannot buy the revolution. You cannot make the revolution. You can only be the revolution. It is in your spirit, or it is nowhere." — Ursula K. Le Guin.

At a birthday dinner not long ago, I posed this question to the friends gathered round: “What has recently cracked you open?”

I was blessed with some mind-expanding answers, but my friend Yondon specifically surprised me by referencing an Ursula K. Le Guin essay, "Science Fiction and Mrs.Brown," a meditation on the inner life of Mrs. Brown, a half-imagined woman on a train, conjured from a chanced encounter with a stranger by the iconic writer Virginia Woolf.

First delivered as a lecture in 1975 and later enshrined in The Language of the Night, Le Guin’s piece draws on Mrs. Brown—Woolf’s emblem for the unquenchable spark of humanity in her seminal essay Mr. Bennett and Mrs. Brown—to interrogate science fiction itself: amid its vast galaxies and ingenious devices, can it still grasp the textured contradictions of a single soul? Or does it surrender the intimate “subject” to mere edifice and idea, forsaking the quivering heart at every tale’s core?

Le Guin wrote:

"I believe that all novels begin with an old lady in the corner opposite. I believe that all novels, that is to say, deal with character, and that it is to express character—not to preach doctrines, sing songs, or celebrate the glories of the British Empire, that the form of the novel, so clumsy, verbose, and undramatic, so rich, elastic, and alive, has been evolved. […] The great novelists have brought us to see whatever they wish us to see through some character. Otherwise, they would not be novelists, but poets, historians, or pamphleteers."

In the haze of that celebratory day, her argument hit differently: the root of great science fiction isn't escapist imagination, but the discipline of observation. Arrakis, Gethen, even the galactic archipelago of Trantor and the Foundation worlds aren't memorable and endurable just because of the spectacle. They persist because their inhabitants pulse with contradiction and texture; every invented custom, political system, or climate is filtered through the nervous system of someone real. The galaxies and spaceships are scaffolding. What sticks is the raw, subjective ache of a person wrestling entropy. The grandeur of the imagined only matters to the extent it sharpens our focus on the subject at its core.

Le Guin's insight is as much a challenge as a comfort. It's easier, especially for founders, for technologists, for anyone trying to pull the future forward, to believe in the mythology of invention: you imagine radically, and then you build. But Woolf, and Le Guin after her, asks: what good are all your objects, your inventions, your universes, if there's no subject at the center? What's a new world worth if Mrs. Brown herself never appears, whole and trembling, on the page?

Today's Mrs. Brown

But what does it really mean, in a world obsessed with invention and speed, to "see" the subject? What does it cost, psychologically, structurally, to resist the economic and technological pressure to build for universality, scale, and abstraction?

Abstraction seduces with its ease. It comforts, beguiles. It spares effort, swells effortlessly, distilling the world’s wild tangle into neat cells for ledgers. As Byung-Chul Han wrote in Non-Things:

"The digital world is a world without smell, without taste, without touch, without pain. It is a world without ritual, without event, without memory. It is a world without friction." This frictionlessness, Han claims, leads to a "crisis of presence," a flattening of experience into sameness.

This indictment cuts straight to the heart of Silicon Valley’s love for detachment and device. Countless ventures erupt with flawless presentations and gleaming trajectories, engineered for some nebulous “vision,” oblivious to the Mrs. Browns hovering at the edges. The lure is spectacle, novelty masquerading as revelation. But transformative tools arise from reverent vigilance toward the already vital and breathing.

Le Guin’s fire burns not in her conjured structures, but in her unyielding devotion to the raw human, the specter that stalks every contrived emptiness.

Founders as sci-fi writers

It's not uncommon to analogize modern founders as sci-fi writers; that's why we frequently see "manifestos" on company sites, a preface of sorts that anchors the prospective readers (the customers and potential hires) to situate essential properties of the world that is about to be introduced. Great founders build great worlds that are nothing short of science fiction. That can obviously be true for Musk, who epitomizes such an archetype, but it can equally be true for the founder of a database company.

At their best, founders mirror Le Guin: patient observers, building on truths gleaned from long study of real needs, idiosyncratic people, local pain. At their worst, they are set designers on a cosmic soundstage, swapping in new props and hoping meaning emerges from machinery alone. This echoes the lament of the talented designer Jason Yuan, who recently observed on X: "what a time: an abundance of capital and tech and optimism .. strangled by a poverty of imagination."

Contrary to common belief, most world-building vision actually stems from a maniacal focus on one specific unit of obsession.

Frank Herbert’s Dune exemplifies such craft. Beneath its colossal worms, flying craft, and cosmic hierarchies pulses a meticulous weave of environment, faith, authority, and lineage. Herbert delved for years into arid biomes, raising his edifice grain by grain. Arrakis’s intrigues circle eternally around water: its dearth, dominion, ceremony, sorrow. Imperial vastness invariably circles back to the particular and palpable.

I think often of a scene in which a founder gestures to a diagram on the deck labeled "Human-in-the-Loop," and I was curious about the actual human in the loop: a worker in Manila, eyes straining at cheap monitors, improvising fixes for a dataset full of contradictions. Who, in the sleekly designed deck, is Mrs. Brown?

The real power of observation isn’t just a moral challenge, but a strategic imperative. Building for the “blurred edge,” Mrs. Browns, the users who contradict any clean-cut persona, is what makes systems resilient. Businesses that survive the volatility of scale are those willing to trace the pattern to its source, even when it leads to the counterintuitive or the overlooked.

The true story of Pangu

In Shenzhen, a man lay sleepless, crushed beneath pilfered lines of code, each a mute accusation, every midnight alert a call from an estranged reality. His laptop, lingering warm at his side, spilled a dim, reproving light, bearing witness to purpose’s slow decay.

His hands, once as steady as code itself, now trembled above the keyboard, twenty-seven years of muscle memory short-circuited by something that had nothing to do with fear of being caught, but the quiet betrayal of one's dignity and purpose.

This is the imagery I conjured in my head when reading the lacerating confession from the whistleblower within Huawei's Pangu foundation model team, a document composed in the dead of the night, tears smudging lines between reminiscence and rage.

"I have been tossing and turning every night and having trouble falling asleep," the anonymous insider confesses. "This kind of repression and humiliation has gradually worn away my feelings for Huawei, making my time here gradually muddle-headed, confused, and often doubting my life and self-worth." He admits, "I am a coward," terrified of losing his job, yet driven to speak.

Pangu, Huawei's audacious bid to forge "the world's second choice" in foundation models, ignited with fervor. Engineers forfeited weekends, family bonds, even months in Suzhou isolation camps, Saturdays defaulted to grueling work, sweetened only by occasional crayfish and afternoon tea, for a cause cloaked in nobility. "We are really burning our youth to polish the domestic computing power base," he writes, evoking the sacrifices: "Living in a foreign city, we gave up our families, vacations, health, entertainment, and sacrificed our lives."

But cracks emerged. Early models, trained on limited computing power, faltered. The 135B, meant as a triumph, was hobbled by a tokenizer so archaic that "each single symbol, number, space, and even Chinese character would occupy a token," wasting resources and yielding poor results. Genuine innovations languished, while the other internal lab’s "breakthroughs" materialized overnight. The revelation: they were essentially cloning competitors. For the new model, they "actually used Qwen 1.5 110B to continue training," adding layers and tweaks to masquerade as original, with parameter distributions almost the same and even the class name of the model code was Qwen at the time, and they were “too lazy to even change the name."

Leadership, possessed of a willful ignorance, nodded sagely at "ten points across all metrics," mistaking forgery for breakthrough. The digital equivalent of art theft, Qwen 1.5's 110B parameters, stretched and disguised, paraded as native genius, was rewarded, while genuine invention was quietly buried.

Now resigning as he writes, he seeks atonement: "Having signed on these technical reports is a stain that I will never be able to erase in my life…if I disappear, it can be considered that I sacrificed myself for the truth and ideals."

Mrs. Brown once again speaks, not as an abstraction. Her voice echoes across borders, a reminder that AI's arms race devours personal ideals indiscriminately, from Suzhou to Silicon Valley. Even in China, often critiqued for systemic invasions of IP and originality, an engineer's sacred dignity in their own work endures, at least within themselves, fueling such a raw and painful confession.

The Stubborn subject

This is what we rarely acknowledge when efficiency becomes the sole plotline, when we hail the falling cost of compute, trace "the end of human operating expense," or build models designed to abstract away labor. The story, as usually told, veers easily into moralizing: technology replacing people, platforms extracting value, Mrs. Brown consigned to a footnote. But there's a subtler danger in centering only loss or only sympathy.

Mrs. Brown is not just a symbol for the displaced. She is also the improviser, the adapter, the one who finds within techno-capitalism new forms of leverage, a way to leap across boundaries, to earn, to transform her trajectory and that of her family. In places where the platform economy or distributed labor offers more freedom than the village or the factory ever could, critique from a distance, especially from a place of privilege, can flatten intent as much as abstraction does.

To truly see Mrs. Brown is not to enshrine her as victim, nor to glorify "resilience" as a feel-good metric. It is to let improvisation and exploitation, constraint and freedom, loss and agency, all tangled together. We miss the point if we only grieve the jobs or celebrate the gains. The only real error is in assuming there is ever just one story to be told.

There's something almost alchemical about the way Mrs. Brown keeps slipping back in — as the anonymous bug report, the unexplained spike in a dashboard, the stubborn edge case that resists automation, the off-color joke that slips past the LLM's filter. Platforms might yearn for labor to become universally fungible, infinitely updatable, and completely replaceable. Yet again and again, the most essential work proves itself stubbornly local, full of contradiction and context, flickering through the cracks of every abstraction, refusing to disappear.

And what of the so-called "builders above the LLM" who wield platforms instead of simply feeding them? Even here, the "subject" erodes: you slide toward becoming an API endpoint in someone else's schema, your edges and voice softly sanded down.

The recursion is relentless: at some layer, in some meeting or migration or late-night call. We are all, at least for a moment, Mrs. Brown.

What good are all the objects?

Every technologist should be haunted by Le Guin's dangerous question: "What good are all the objects in the universe if there is no subject?"

It's easy to analyze geopolitics and technological arms races at the level of nations and models, to speak in terms of "capabilities" and "regulatory arbitrage." It's infinitely more complex, and as urgent, to tell the stories of subjects: individuals with pride and shame, with families they hope to protect, with secrets they never meant to leak.

Why is this inversion so hard? Ironically, the local resists measurement. Mrs. Brown rarely shows up in the metrics that drive a product roadmap or a national GDP. To see her, one needs slow attention. One needs the discipline to sit in the complexity, to resist the drive to "close the ticket," "satisfy the metric," or "ship the feature" by rounding away the rough contour of the subject.

We keep trying to engineer Mrs. Brown away, and every new world we build, her shadow grows larger. The more frictionless our systems become, the more haunting her absence feels. Friction, interruption, error: these are not always signs that the system failed, but signals that the subject is still in the room.

Maybe the only real test of a new world, technological, narrative, or otherwise, is whether Mrs. Brown feels real enough to write a confession that cracks the system open. The future, whenever it arrives, will be built from the inside out. And suppose those creating it aren't allowed to remain human, to feel pride, shame, exhaustion, and the courage to speak truth. In that case, it will collapse under its own frictionless weight, leaving only mechanical abstractions and the ghosts of the subjects it consumed.

In the end, as Le Guin knew, the revolution begins not with code or conquest, but with the quiet act of seeing of being the subject who trembles and demands to be seen.

Note: For those seeking the full context and texture of the whistleblower’s confession, I encourage reading the original document. Consider using your browser’s translation feature or an LLM to access the testimony in its own words.

A friend recently published her first book. Despite critical praise from her immediate literary circles, her publisher delivered a stark verdict along the lines of having to decide whether to go for the “National Book Award” or for “BookTok.” The literary middle market is dead.

The Barbell theory originated as a portfolio diversification strategy in hedge funds, popularized by Nassim Taleb. So it’s cool to hear it in the zeitgeist. Barbell theory is eating the world (or at least, the timeline). I’d be having a conversation with someone about the return to traditional values. The barbell theory would suggest that you either optimize for the hyper-capitalist, profit-maximizing, “secure-the-bag” Machiavellian, or pure, good-hearted, God-abiding family life. You gotta barbell it.

All of a sudden, the theory seems to have become the dominant metaphysical framework to understand what’s going on. It posits that technological acceleration (e.g., algorithms) and network effects have fundamentally altered the distribution of outcomes across all domains of human achievement. Where bell curves once dominated, describing a world of averages and incremental gains, we now see radical bifurcation: a world splitting into exponential winners and absolute losers, with the middle ground not just shrinking but actively hostile to value creation.

In many ways, barbell theory is the inverted bell curve, just like this diagram by Peter Thiel at his lecture at Stanford.

The metaphor itself is seductive in its simplicity: a weightlifting barbell, with heavy loads at both ends, and a thin connecting bar in between. No weight in the middle. No value in the average. It's a compelling image for our winner-take-all zeitgeist.

This isn't merely about income inequality or market dynamics. The barbell theory has become a kind of secular eschatology, promising salvation through extreme outcomes while condemning the middle path as a form of living death. Tyler Cowen's "Average Is Over," the book that provides some data points to this philosophy, argues that technology will cleave humanity into a cognitive elite who master technological leverage and a vast underclass subsisting on its outputs.

However, the framework, while compelling, confuses a narrative convenience with a law of nature. And even more interestingly, the belief of the presence of the barbell theory reflexively enhances the barbell nature of any situation, even if the outcome of converging to either side of the barbell rationally yields a worse outcome than simply being “average.”

I call this shaded area ‘voluntary loss,’ the opportunity cost we accept when we force a world of nuance into a binary frame, and in doing so, forfeit the steady gains of the middle. Admittedly, this is a reductive abstraction, intended solely to illuminate the underlying principle.

Barbell theory, compelling and pervasive as it is, has begun to shape not just our perceptions but our actions, nudging us inexorably toward its extremes. We chase the alluring ends, convinced that mediocrity or average pursuits carry no expected value, only silent stagnation. NO ONE WANTS TO BE MID.

The barbell’s gravitational pull distorts everything from career choices to patterns of consumption. College graduates toggle between “grinding LeetCode (or cheating) for FAANG” or “getting into YC,” dismissing stable mid-tier opportunities as invisible or irrelevant. Creative professionals oscillate between chasing viral fame and risking obscurity, while the steady, cumulative path of craftsmanship fades from view. Writers face the same false choice my friend encountered: literary prestige or viral BookTok fame, as if building a readership through book clubs, independent bookstores, and word-of-mouth had somehow ceased to exist.

Even consumer preferences seem to be bifurcating: people seek either the curated luxury experience or the no-frills roadside dive, bypassing the everyday reliability of the neighborhood restaurant, the local bookstore, the mid-tier hotel. These casualties of the barbell worldview illustrate how the value of the middle is systematically erased… but is that really our revealed preference?

Let’s say we apply the Barbell theory to this somewhat silly example of choosing a restaurant: either a Michelin-starred restaurant or McDonald’s, with nothing in between. It’s probably quite obvious that it’s not always true. Picking a more “normal” restaurant probably wins out in both value and quality compared to either the Michelin-starred option or fast food.

It is worth pausing to delineate the mathematical contours of this difference:

• A normal distribution curve represents scenarios where most outcomes cluster around the average, with extreme outcomes becoming increasingly rare. In such a model, incremental improvements yield significant EV because moderate gains are highly probable.

• The barbell theory proposes an inverted or bifurcated distribution, where outcomes are concentrated at the extremes, and middle-ground outcomes are comparatively rare or even considered negligible.

Critically, assuming a barbell-shaped reality in situations that more accurately reflect a normal distribution can lead to suboptimal decisions. If moderate success is indeed common and valuable, ignoring it for a slim chance at extremes is mathematically irrational and practically disadvantageous.

In the startup landscape, this distinction is starkly visible. Venture capital's devotion to power-law returns fosters the illusion of binary outcomes. Founders are sometimes compelled to pursue unicorn status or accept oblivion. Yet the practical reality is a bit more nuanced.

In venture, returns cluster at extremes, algorithms drive engagement to poles, and as Peter Thiel notes, founder personalities that seem unhinged in normal life often yield extraordinary outcomes. These power laws are real. But venture capital's outsized cultural influence filtered through TechCrunch headlines and YC playbooks has led to the overgeneralization of what should remain a specialized model. Not every business needs to be a unicorn or a bootstrap. Not every domain follows power law distributions. Yet everyone increasingly feels somewhat pressured to choose between these extremes, as if the middle ground were mathematically forbidden rather than merely unfashionable.

This distortion extends beyond individual choices to our collective decision-making, such as our political discourse. Society increasingly perceives policy debates as stark binaries-socialism versus capitalism, liberty versus control. Yet actual policy outcomes typically conform more closely to normal distributions, with incremental improvements and moderate policies offering meaningful expected value, enhancing overall societal well-being.

Barbell theory is not merely a description of the world; it is an incantation that, once believed, reflexively shapes reality in its image. The more we insist that only extremes matter, the more we engineer systems—investment structures, cultural algorithms, even personal ambitions—that reward only the outliers and starve the center. In this way, the myth becomes self-fulfilling: the middle ground is not lost to entropy, but actively dismantled by our collective refusal to see its worth.

Yet beneath this manufactured scarcity, the middle is not a wasteland of mediocrity but the quiet engine of civilization itself. It is the unglamorous store manager, the midlist novelist, the regional manufacturer—those who persist in the steady work of maintenance and incremental progress. They are the ballast that keeps the ship upright while the spotlight chases the acrobats at either end. The barbell’s true paradox is this: we chase the extremes, but it is the center that endures, adapts, and ultimately sustains the system’s very possibility.

The middle isn't dead; it's just harder to see beneath the glare of extreme outcomes.

Perhaps barbell thinking is itself a historical anomaly, the intellectual artifact of a very specific moment in time. The venture boom of the 2010s, fueled by zero interest rates and platform monopolies, created a generation that mistakes temporary conditions for eternal truths. When money was free and winner-take-all dynamics seemed inevitable, extreme outcomes appeared to be the only rational strategy.

But what happens when the underlying conditions shift? When interest rates rise and capital becomes expensive again? When does antitrust action fragment platform monopolies? Network effects and attention capture endure, but their contours shift. Influence pools in unexpected corners. New gatekeepers emerge—platforms, personalities, algorithms—redrawing the boundaries of power. The barbell world was built on a foundation of cheap money, amplifying specific types of network effects. As those financial conditions change, the same underlying forces may reward different strategies.

The evidence points in two directions. We may be witnessing a permanent structural shift, where technology has irreversibly concentrated power and returns. We can also believe that we’re in a particular phase of a longer cycle. The post-war boom of suburban businesses, regional newspapers, and mid-market companies was once considered the "golden age" of moderate success, coinciding with regulated markets and geographically constrained competition. The current barbell era emerged when global digital networks removed those constraints. But regulation and competition patterns are already shifting again.

What is certain is that the assumption that moderation equals mediocrity is historically false. Warren Buffett did not build his empire through reckless moonshots, but through the patient, almost monastic ritual of capital allocation. Sam Walton created the world's largest retailer by perfecting the unglamorous middle market, store by store. These weren't small ambitions executed moderately, but massive ambitions executed with methodical discipline.

The founders building quietly profitable businesses today may not be missing the future. They may be building it. They built their empires one customer, one store, one reader at a time, the same steady accumulation that publishers now tell authors no longer exists.

I'm (re)joining Pace Capital, a firm that invests in and develops pre-consensus ideas.

Unprecedented times call for unprecedented partners.

We're building a temple dedicated to clarity and conviction, forming ruthlessly honest partnerships with obsessives pursuing their life's work.

You can view the interactive version of the announcement here.

Every morning, we wake into histories unfolding, currents of possibility quietly reshaping the future. We stand together at the edge of a great transformation, sensing the gathering force of change. This moment is thrilling and unsettling, stripping away distractions and revealing the one thing that truly matters: our infinite capacity for creation.

Open-source chips, on-device inference, agentic systems, ambient interfaces, generative media engines, and 24/7 onchain markets have scattered the seeds of genesis into every corner of our lives. We have compressed the gap between imagination and instantiation down to milliseconds.

Such profound compression demands a new kind of partner. Partners who can navigate entropy, sense the right moment to intervene, and place contrarian bets before the faint signal becomes consensus. Partners possessed by the same obsession that keeps founders awake at 3 a.m.

Obsession isn't a lifestyle, it's the necessary cost of conviction. I learned that building my startup Station Labs, hammering late‑night builds with my co-founders until we finally felt like the product was pushing the boundaries of onchain DevX and UX. We later got acquired by Coinbase to become part of Base, where I led the 0 to 1 of some of the most well-loved developer tools including OnchainKit, MiniKit, and Base MCP. Obsession compels us to stake a claim long before others even realize there's a map.

The most transformative opportunities arise where deep obsession meets potent societal shifts. The migration from traditional finance to crypto rails, the transition from fossil fuels to renewables, the decentralization of inference to the edge, and the evolution from passive media to personalized experiences—all begin subtly, then swiftly remake our world.

Building at these turning points, just before their magnitude becomes undeniable, fuels my day and keeps me awake at night.

I've known the Pace team and witnessed its growth for many years, first as a junior investor, then as a founder they backed. As much as rejoining the team feels like homecoming, it feels more like joining something brand new: bigger, sharper, and even more assured of what it represents in the ecosystem.

Pace is built for the sliver of time after impossible and before obvious; we practice pre‑consensus investing: go deep, decide quickly, ignore the crowd. We trade broad opinions for sharp theses, comfort for honesty, gradients for stark clarity. Exceptional investing at Pace transcends identifying great companies; it demands precision of thought, rapid learning curves, and dedication to founders.

Central to our philosophy is "double-loop learning", a perpetual refinement of frameworks that illuminate not only the essence of transformative businesses and technologies but also our deepest intuitions and biases. This discipline compels us beyond the question, "Why does this company matter?" and into deeper territory: "Why does it resonate uniquely with me, and what makes me the most valuable partner to this founder?"

Double-loop learning inherently rejects superficial analysis or detached market maps that miss the essence of extraordinary, category-defining companies. Instead, we strive for profoundly personal, fiercely introspective clarity, identifying the sparks destined to ignite new markets before they're visible to the world.

While others construct empires, Pace deliberately wants to build a temple dedicated to clarity, originality, and conviction. Our ambition isn't measured merely by the size of the funds we raise, but by the magnitude of returns our clarity unlocks, the audacity of our challenges to orthodoxy, and our courage to act decisively before consensus emerges.

Authentic partnerships here with founders are born not from comfort but from relentless honesty. We rigorously interrogate assumptions, sharpen intuitions, and demand intentionality at every juncture. We commit ourselves to visionary founders whose obsessions align with the most potent undercurrents reshaping society, whose work defines not just markets but entire paradigms.

If you are tugging at the seams of reality, writing the first line of code, a half‑finished memo, a prototype repo, a voice note scribbled at 3 a.m.—I will meet you there, in the dark, before the world wakes up. I publish ideas, theses, and code frequently, and will be a skin-in-the-game partner in crime. Together, we can make inevitability arrive sooner.

Let's build.

Note: I will continue to write on this blog consistently, as writing is thinking, and the clarity of thoughts can only emerge through the discipline of expression. This space is not just an announcement board, but a living document of the questions, frameworks, and insights that shape my work at Pace.

Thank you for being a part of it.

When SpaceX positions itself as humanity's salvation, it taps into something deeper than progress or profit: a hunger for the sacred.

Silicon Valley's power didn’t emerge from code or capital alone. It emerged from the collective capacity to experience technology as sacred, as something that transcends mere utility to touch deeper human aspirations.

However, when mythology meets market incentives, something shifts. The same framework that elevates founders into visionaries and startups into movements becomes a sophisticated performance art. In the attention economy, the appearance of innovation often yields better returns than innovation itself.

This tension between authentic sacredness, which drives genuine breakthroughs, and manufactured sacredness, which enables sophisticated fraud, threatens not just individual companies but the entire ecosystem's capacity to create meaningful progress.

This loss of sacredness haunted Mircea Eliade when he warned of the "total profanation" of human experience in "The Sacred and the Profane," a book I've carried with me through 2025. In Eliade's view, modern society's greatest peril wasn't secularization but our diminishing ability to recognize the sacred in anything. For homo religiosus, certain spaces, objects, and events transcend mere function, becoming vessels of deeper meaning through what he called hierophanies.

Eliade's framework helps explain Silicon Valley's unique cultural power. Silicon Valley is built on this very tension between the sacred and the profane, between what can be measured and what can't, between the explicit and the implicit (or Straussian).

Engineers are as passionate about "mission" as monks are. The stories that founders tell about where they came from are like myths. Product launches are set up like hierophanies, which are ways of showing the sacred in everyday life.

Yet the same forces that create this sacred dimension also threaten to destroy it. When every interaction becomes optimizable, when every ritual is reduced to its functional components, we risk creating a world stripped of transcendent meaning, yet also fulfilling humanity's deep wish to be relieved of the burden of mystery.

I find myself asking: Is it still possible to preserve sacredness?

The sacred foundation

This sacred dimension manifests most clearly in how the Valley treats its founders as mythological figures whose personal journeys embody broader technological transformation.

As Peter Thiel observed in his famous Stanford lecture, founders occupy a peculiar place in the Valley's collective psyche, simultaneously worshipped and vulnerable to sacrifice. They are extreme insiders by their power and vision, yet often extreme outsiders in their eccentricities and willingness to break convention. The result is a strange hybrid: manufactured mythology running on engagement metrics, where founders must navigate between being deified and demonized.

The Valley's founding stories often follow ancient mythological patterns: the outcast-turned-hero (Jobs), the boy-king (Zuckerberg), the Prometheus figure bringing divine fire to mortals (Musk). Each faces trials, betrayals, exile, and potential restoration. As Thiel notes, this isn't merely a metaphor but a fundamental dynamic of how power and narrative operate during times of transformation.

What does an investor look for when evaluating a founder? The best investors have developed an almost mystical framework, seeking three critical signals in founders: the capacity for transformation, the willingness to face ordeal, and the potential for resurrection.

For a founder, like an ancient initiate, the journey demands a readiness to shed their former self and embrace rebirth. This transformative process is marked by specific indicators: their ability to relinquish ingrained assumptions, their resilience in navigating prolonged uncertainty, and their capacity to motivate others to venture into the unknown alongside them.

Investors intensely focus on a founder's personal narrative due to identifiable mythological patterns. As Thiel observes, the evolution of founders into sacred figures follows a consistent trajectory: the initial blessing (funding), a period of adversity (the "valley of death"), and eventual triumph or failure. These stages align with key investment milestones and resonate deeply with ancient initiation rituals.

When founders succeed at scale, their companies can transcend mere business to become something approaching cosmic centers. When certain companies achieve "escape velocity," they genuinely become axis mundi, cosmic center points around which entire ecosystems orbit. Consider how Anduril's American Dynamism narrative has catalyzed an entire movement, reshaping conservative tech discourse and defense innovation. Or how OpenAI's alumni have spawned distinct philosophical schools of thought about AI safety and ethics. These organizations don't just build products; they restructure how entire communities think about technology's role in society.

However, a crucial distinction lies between authentic and manufactured sacredness. While both involve the creation of sacredness, there's a key distinction between its organic development and its manufactured imitation. Traditional sacred spaces acquired meaning gradually over time, whereas contemporary platforms strive to hasten this process, fabricating significance through growth hacks and viral loops.

As Thiel observes, this acceleration creates a dangerous instability: the faster you build sacred power, the more vulnerable you become to sacrificial dynamics.

The profane intervention

True sacredness is rooted in forms of value that cannot be bought or sold. This is precisely what makes it so vulnerable to exploitation. The Valley’s most successful founders intuitively grasp this paradox. They must build financial value while tending to what might be called “sacred capital.” These are measurable only in their aftereffects—retention rates, organic growth, the gravitational pull on talent and discourse—but their essence is irreducible.

Open-source software contributors, Discord servers, or Subreddit mods operate on the principles of gift economics. The most vibrant online spaces run on similar dynamics, where value flows through reputation and contribution rather than transaction. There's a reason we recoil when someone tries to reduce these relationships to a score or metric. They lose their essence when abstracted from their living social context.

Modern organizations need metrics to operate, but meaning to thrive. The best founders learned to walk this knife's edge: to honor both the ledger and the ineffable, to tend simultaneously to what can be counted and what must remain untamed.

Yet this same sacred foundation that enables Silicon Valley's most transformative innovations also creates vulnerabilities that bad actors increasingly exploit. The crypto space serves as a concentrated laboratory for understanding how sacred narratives can be corrupted into sophisticated fraud.

No industry dances more closely between the sacred and profane than crypto. The utility that we see now – USDC, DeFi, and a faster blockchain itself – only emerges after the fervor of an early belief that individuals are sovereign, and that permissionless systems will reclaim power from institutions and return it to the people. It’s as much a financial or technological movement as it is a cultural-political one.

When we reduce everything to tradeable narratives, we inadvertently create the conditions for fraud. I've watched brilliant founders start with genuine technological innovation, only to get caught in an escalating spiral of story over substance.

The unrelenting demand for captivating stories in the market can easily disorient even the most well-intentioned teams. Every exaggeration is excused by the need to sustain progress.

The worst fraud isn’t even the loss of capital, but the betrayal of faith. It erodes the very foundation of trust that makes the Silicon Valley miracle possible in the first place.

The line between myth-making and deception is treacherously thin. While every founder must craft narratives that inspire belief, recent years have shown how easily sacred storytelling can descend into outright fraud. Everyone in tech knows the cautionary tales of SBF, Elizabeth Holmes, and Adam Neumann—each a master, in their prime, of bending reality to their will.

A new crop of founders is now emerging who understand that in the attention economy, the sacred isn't earned through genuine breakthrough or sacrifice, but manufactured through narrative. For them, the work of building an honest business feels relatively inefficient compared to crafting a compelling hero's journey.

Their playbook inverts Eliade's hierarchy: instead of technology transcending its mundane function through genuine innovation, they aim to elevate the mundane through pure storytelling. "Tweeting one's way into every room" becomes their hierophany, even at the risk of going to jail for securities fraud.

The incentives perfectly reward this inversion of sacred and profane. Why toil in the profane realm of customer interviews when you could be crafting your David versus Goliath mythology? Why chase the mundane validation of product-market fit when a viral thread about regulatory persecution could instantly consecrate you as a visionary? The attention economy has created a new priesthood, where the sacred rite of "founding" requires no actual founding, only the ability to perform the outward signs of “founding.” A provocative take that catches Paul Graham's eye feels more transcendent than the quiet progress of serving customers. The metrics that matter aren't profane measures like retention and revenue, but sacred tokens of narrative resonance: retweets, threads gone viral, the digital liturgy of startup culture.

Well, but what do they say? “First-time founders are obsessed with products, second-time founders are obsessed with distribution.”

Why don’t I front-run sone hard-earned lessons?

The cost of denial

The systematic denial (or betrayal) of Silicon Valley's sacred foundations threatens not only individual companies but also the entire ecosystem's capacity for genuine innovation. When founders exploit the delicate, implicit trust for their own deception and short-term valuation inflation, they diminish the collective faith that is the lifeblood of bold technological progress.

When a critical mass of sacred narratives is exposed as cynical manipulations, the ability to discern genuine visionaries diminishes. Investors, in turn, become increasingly wary of transformative assertions. This erodes the gift economy, where knowledge, mentorship, and collaborative innovation are freely shared, as participants become more defensive and transactional.

The feedback loop is devastating.

The dynamic was most apparent recently in crypto, where projects start with genuine technological innovation, only to discover that engineering excellence means nothing without a "narrative fit." The pressure to maintain token prices and community momentum gradually hollows out the sacred act of building something meaningful.

Efficient mechanics drive rapid growth. A founder's tweet generates Discord channels; a meme launches a valuable token. The "community" is both customer and product, incentivized to spread the message, labeling critics as "non-believers." Innovation transforms into sophisticated belief propagation.

In some strategy sessions, I watched teams debate not technical architectures but narrative arcs. “What’s our story?” gradually overshadowed the more elemental question: “What are we building?”

Projects with solid fundamentals have a harder time breaking through the noise, while those running purely on vibes and clever tokenomics soared. Each successful manipulation made the next one easier, until the space became nearly impossible to navigate without participating in the shared delusion.

Young founders, watching this dynamic unfold, internalize the lesson that sacred narratives are merely instruments of manipulation rather than genuine repositories of meaning and purpose.

The very idea of building something truly sacred, something that transcends pure financial calculation, begins to feel naive rather than noble.

Preserving sacredness

This essay is less a prescription than a reckoning, a meditation on what we stand to lose. Perhaps the point isn’t to prescribe fixes, as the profane always coexist alongside the sacred, but to heighten awareness of how our choices, as creators and stewards of technology, shape the incentives that define the soul of our industry.

The sacred isn’t some vague notion of a “soul of tech.” It’s the spark in a teenager’s eyes when they realize they can shape the world with code. It’s the midnight breakthrough that feels like catching lightning in your hands. It’s the force that transforms this industry from mere business into something that feels, at its best, like magic.

The good news is that I still see sacredness everywhere. A relentless focus on the work yields a superior outcome. There’s something about seeing good people win that reminds you that people do notice.

“The opposite of slop is care.”

Said Nabeel S. Qureshi on the Dialectic podcast. Moments of sacredness are moments of devotion, the faithful pursuit of excellence, because our internal compass points there.

I maintain this blog as a small ritual of preservation. Each post is an attempt to capture those moments when technology transcends its utility and touches something that’s deceivingly simple about being human: the wonder, the discovery, the desire for truth, and the impulse to ask fundamental questions.

The sacred will persist because we’re willing to defy its ossified forms and create brand new vessels capable of holding its essence.

The sacred is essential to who we are as creators. In protecting it, we protect our ability to dream.

A student asked master Joshu, "What is Buddha?"

Joshu answered, "Three pounds of flax."

The student was enlightened.

The merchant who overheard this doubled his flax prices.

I was having coffee with a founder last week when he started talking about AI-enabled services business roll-ups. His eyes lit up as he described the opportunity and wished he could pivot from his current work to pursue it. High-profile fundraises, influx of capital, and the chance to consolidate fragmented service industries using AI.

I asked him to clarify the specific role of AI within the proposed strategy. Upon further discussion, it became evident that AI was not fundamentally altering the capital structure or unit economics of these businesses. The primary justification centered on reducing operating costs. Historically, the success of roll-up companies has depended not solely on software, but on access to talent, capital, and the capacity to underwrite acquisitions more effectively than competitors.

My friend, the founder, spoke without irony or guile; belief is an insidious comfort. AI had become his talisman, a glittering promise to guard against the terrors of irrelevance. But the system rewarded AI narratives over the mundanities of debt capacity, third-party integrations, and operational efficiency.

We live in a moment when the letters ‘AI’ radiate with almost mystical power, conjuring illusions that obscure the machinery grinding beneath. Like all magical incantations before it, the phrase ‘artificial intelligence’ hides more than it reveals. Investors whisper it with awe and greed; founders chant it like a secular prayer. Yet the machine itself remains unchanged. This is what the market fervently hopes we do not notice.

Byrne Hobart’s newsletter, The Diff, attracts finance and technology audiences by critically examining and clarifying the complexities of markets and systems.

For example, while Hobart may address American college admissions, his analysis focuses on the underlying mechanisms of elite-selection systems. He examines how institutions compress talent signals, delegate vetting through hierarchical structures, and encounter strategic manipulation as processes become more transparent. In discussing Archegos, he elucidates how concentrated leverage internalizes risk, resulting in significant failures among a small group of participants while preventing broader systemic consequences.

Whether he names it or not, Hobart practices Systemantics, the discipline of noticing what a system really optimizes for.

And the one simple goal of any student of Systemantics is to answer the question: what’s really going on here?

Three axioms of Systemanics

In 1975, John Gall published Systemantics: The Systems Bible, a book that should be required reading for anyone wanting to build or understand systems. Gall identified dozens of axioms throughout the book, many hilariously simple. I found these three fundamental truths most relevant:

THE NAME IS MOST EMPHATICALLY NOT THE THING.

Labels set expectations reality rarely satisfies. Say “restaurant,” and visions of cuisine appear. Say “venture capital,” and money springs to mind. Both mislead subtly and profoundly.

PEOPLE IN SYSTEMS DO NOT DO WHAT THE SYSTEM SAYS THEY ARE DOING.

University professors often prioritize securing grants over fostering student learning. Venture capitalists focus on building personal brands in addition to managing investments. Startup founders frequently optimize for funding metrics rather than addressing customer needs.

THE SYSTEM ITSELF DOES NOT DO WHAT IT SAYS IT IS DOING.

The restaurant industry often generates most of its profits from alcohol sales rather than food. Similarly, the printer industry derives significant revenue from ink cartridges. Reasoning models may not perform genuine reasoning.

Although these principles may seem cynical or pessimistic, they accurately reflect practical realities. Systems typically evolve to prioritize their own sustainability over the functions suggested by their names.

The pattern everywhere

Once this pattern reveals itself, it manifests everywhere.

Gas stations present themselves as fuel providers, yet their primary profits derive from high-margin snack sales. Gyms promote personal transformation but rely financially on members who do not regularly attend. Restaurants emphasize culinary experiences, though their profit margins often depend on the markup of alcoholic beverages.

Many successful companies generate significant revenue from sources rarely highlighted in their marketing materials.

DoorDash operates as a food delivery platform but also functions as an advertising network. Restaurants not only pay commission fees for filling orders but also increasingly invest in promoted placements, sponsored listings, and customer acquisition tools. While the delivery infrastructure aggregates demand-side customers, high-margin advertising products contribute significantly to the bottom line.

We all know Nordstrom as a high-end retailer, but its financial stability is driven more by co-branded credit card partnerships than by apparel sales, especially during off-seasons. Apparel typically operates with minimal profit margins and significant inventory risk. Co-branded credit cards, on the other hand, offer contribution margins of 90%+ or more, require no physical inventory, and continue to generate revenue even during periods of low store traffic.

Robinhood offers commission-free trades to retail investors, but its primary revenue streams are the sale of order flow to HFT firms and the interest earned on consumer cash deposits. While free trades attract users, payment for order flow constitutes the main source of revenue.

Looking at things through the lens of Systemantics helps explain why tech hype often hides the real economics underneath. Usually, new technologies are added to existing ways of making money, so business models change only a little rather than a lot. Incremental changes rarely build venture-scale outcomes.

Sit in that gap and the real question stops being whether AI works. It becomes who pays for it, and on whose books the savings land.

For example, the efficiency is real because AI can replace some jobs and lower operating costs. But the money that leaves payroll doesn’t automatically become savings, but gets reinvested into tools. These AI tools are paid for based on usage, which grows with the number of tokens used. Some companies end up spending more on inference than they ever did on the people AI replaced. The cost shifted from a fixed expense to a variable one that increases with every user action. The efficiency gain and the higher spending are now two sides of the same coin. If you follow the money, you’ll see it doesn’t stay with the company that “adopted AI,” but ends up as revenue for the company selling the tokens.

There’s a reason people joke that just mentioning AI in your pitch can 10x your valuation. The market is betting on future savings more than the savings that actually exist today. Both investors and founders have reasons to keep this story alive. Investors need a good story to invest, and founders need it to raise money. Neither side is rewarded for noticing that the costs have just changed hands, and this will continue until the token bill or a negative gross margin finally begs a deeper look.

As a founder or an investor, times like these present unique opportunities to win by simply understanding what’s going on.

Why Request for Startups fails

Systemantics explains why top-down ecosystem initiatives rarely produce generational companies.

When YC publishes a Request for Startups, the name of the quest pre-compresses the solution space, proving even Silicon Valley’s idea factory can’t escape Gall’s first axiom.

Naming a desired outcome creates a frame that constrains thinking. When you ask for “a better way to manage healthcare data with AI,” you get solutions that fit that frame. Sometimes, plausible-sounding ideas are the most dangerous because the addressable market is so big that you can put many zeros on a deck without flinching. But those ideas are challenging precisely because the market is large and entrenched, with existing stakeholders and value chains reinforcing the status quo, and incremental improvements are unlikely to yield venture-scale outcomes.

The most transformative companies rarely emerge from someone else’s problem statement. They come from either a steadfast vision, manifested and executed over many years (Steve Jobs with Apple, Elon with SpaceX), or tinkering on the edge from a typically obsessive founder who “stumbles” upon a gold mine after hundreds of trial and error, and the latter can drastically underestimate how big a brand new category can get.

Facebook wasn’t created to answer a call for “improved college social networking.” Instead, it grew out of Mark Zuckerberg’s sense of how Harvard students wanted to connect. Sites like MySpace and Friendster were popular, but they didn’t link to real identities. People liked seeing information about people they actually knew. What started as a better way to look up classmates became a system for mapping identity, attention, relationships, and behavior, which turned out to have tremendous commercial value.

Google wasn’t a business plan to make web search an auction house; in fact, its business model came years later. It started as a curiosity from Page and Brin’s research around applying the logic of academic publishing to the internet where citations equal authority. Their recognized the gap between what search engines claimed to deliver (relevant results) and what they actually provided (keyword matching with minimal quality control), and thought a new algorithm that evaluated both the quantity and the quality of links pointing to a page would help. It did, and became PageRank.

Follow the money

If you want to see through the hype, pay attention to where the profits go, not just what companies claim. Track the incentives along the value chain, and watch for differences between how a company operates and how it markets itself.

One of the hottest new categories is AI [insert existing job functions]. Many heavily marketed “AI SaaS” products are just thin layers that help route prompts to model providers. Their business model relies on arbitrage customer acquisition: using VC funding to acquire users through glmarous marketing, taking a spread from API calls to the model providers, and hoping to build enough workflow lock in or brand recognition before the main model offers the same feature directly. If they aren’t just model wrappers, they are often IT consulting firms where engineers step in to meet service agreements behind a so-called “magic” interface.

CoreWeave is another example. It is a fast-growing “neo cloud” company focused on providing GPUs. However, its most interesting product is not just computing power, but its financing model. The company buys rare Nvidia GPUs, assigns them to a few large customers, borrows money using the hardware and contracts as collateral, and then sells the public a stake in future AI capacity. In 2024, most of its revenue came from Microsoft, while Nvidia acted as supplier, investor, and customer. While the business is marketed as flexible cloud access, the real operation involves GPUs, a few big customers, private loans, leases, depreciation, and debt payments—essentially, AI demand becomes the collateral that supports the whole business. Someone aptly calls it a $22B gamble.

Scale AI says it’s an AI data platform, but it works more like a staffing agency. Human annotators earn up to $8 per hour to label data, while Scale sells this labeled data for up to $100 per image. The company profits from each task by charging more than it pays for labor. The labor didn’t disappear but became hidden, split into small tasks, managed through platforms, and sold as intelligence.

Not every system has a gap between what it says and what it does. Many AI tools really do deliver on their promises of improved productivity and unleash new forms. of creativity. The key takeaway is that it is important to separate this from the business models built around it, which often follow familiar economic logic.

The art of understanding what’s going on

Systemantics, or the art of understanding what’s going on, means recognizing the persistent gaps between what systems proclaim and what they actually do, and capitalizing on that insight.

When the excitement tempers, the people who took the time to understand how things really work will be the ones who last.

Lasting AI companies will appear in two main areas: practical but important tools that clearly improve profits or cut costs, and real frontier research that discovers new problems to solve. The first group improves what already exists, while the second creates what does not exist yet.

Real opportunities lie in the quiet spaces between stated ambitions and operational truths. Just as they always have.

Last week, I came across a confessionary note on the app RedNote, the Chinese equivalent of Instagram and Pinterest. The post was written by an anonymous Chinese VC, whose perspectives were honest, resonant, and refreshing, standing out in a pool of vapid content.

I thought it was so great that I decided to translate the post.

A screenshot of the original post

Translation

Title: "The VC market under my eyes feels heartbreaking 😣"

Of course, many outstanding peers and many excellent companies are emerging. Yet, this market is unprecedentedly subdued, disoriented, and internally inconsistent.

As someone who's been interacting with private equity and venture capital for a decade, I find this deeply disheartening.

Originally, PE and VC were not part of traditional finance—they capitalized on intellectual arbitrage, demanding extremely high comprehensive qualities from practitioners, far beyond mere capital multiplication. But now, intellect doesn't matter, luck doesn't matter, and even the basic notion of money making more money feels irrelevant.

Everyone is busy proving that companies are engaged in "hard tech" with substantial fixed investments, then navigating complex relationships with various government bodies. If a state-owned entity decides to invest, the investment firm follows suit.

In this scenario, "hard tech," large fixed investments, and various forms of state-owned capital all deviate from the original essence of venture capital.

Alienated Technology

Sometimes I wonder if we've confused weight for significance. Why do we equate ‘hard tech’ with heavy machinery, as if real innovation must be physically massive?

Manufacturing is undoubtedly crucial, but investing in manufacturing for the sake of manufacturing alone often leads professionals to overlook financial models.

Moreover, for quite some time, startups founded by scientists have been highly sought after, as the term "scientist" is easily associated with "technology." Yet, ventures led by scientists often end in chaos.

Forced Fixed Investments

Typically, investors (even those buying stocks) prefer companies with lighter assets, especially during economic downturns when cash itself is scarce, and fixed assets sit depreciating. Now, however, fixed investment has become a rigid evaluation standard, mainly because active funds are mostly government-backed industry funds. Entrepreneurs, in their desperation for financing, artificially inflate their fixed investment requirements. Some might present staged investment plans, while others might even fabricate the amounts spent on fixed assets.

Open Relationships among State-Owned Capital and General Partners (GPs)

Nowadays, it's common practice first to identify a project and then seek state-owned capital. State-owned entities often collaborate simultaneously with multiple General Partners, and GPs may present the same project to multiple state-owned entities at the same time. Many GPs have become mere intermediaries—simple matchmakers.

Due to poor market conditions, it's initially difficult to distinguish between truly professional investors and mere intermediaries. Sometimes, these intermediary-type investors, adept at exaggeration and flattery, are even more popular than professional investors.

What pains me most is seeing how those who pursue quick gains (looking for immediate profits) often succeed more easily than those with true professionalism, whether among state-owned entities, entrepreneurs, or investors.

I've even received three different business plans from the same company, each with entirely different valuations, funding needs, and paths to going public, each tailored for different types of investment firms.

Ah, truly disheartening.

Even the two most iconic figures in the Chinese venture capital market have long since left the country.

China's venture capital landscape has undergone a significant shift, with government guidance funds (GGFs) now dominating and a growing focus on "hard tech" following a tech crackdown. Private deal volume and value significantly dropped between 2022 and 2023. Traditional innovation sectors, such as consumer social, education, and gaming, have declined, giving way to advanced manufacturing and fields like artificial intelligence. The government has implemented measures to address issues with guidance funds, forcing Chinese VCs to prioritize national goals over market opportunities, thereby reshaping China’s innovation ecosystem.

The confession is a warning of innovation degenerating into spectacle, where optics eclipses substance. Investors become matchmakers, companies present tailored narratives, and "hard tech" is idolized regardless of viability, symptoms marking the late stage of the “spectacle of building.”

The spectacle of “building" describes systems that optimize for whatever can be ritualized, narrated, and scaled. The argument isn’t that spectacle alone is negative. Quite the opposite. Byrne Hobart and Tobias Huber, in their recent book Boom, make the persuasive case that bubbles aren't merely collective irrationality; they’re crucial for innovation. Bubbles draw in the capital, attention, and human energy needed to chase ambitious, improbable ideas.

Most ventures fueled by these bubbles implode dramatically. But at the scale of entire ecosystems, this frenzy of experimentation accelerates discovery, pruning countless bad ideas while swiftly elevating the few that truly matter. Without speculative exuberance, railroads, the internet, or even AI might never have reached escape velocity.

The crucial difference isn't whether spectacle exists, but how quickly and openly these collective delusions respond when reality reasserts itself.

With the new administration in America, we're also seeing a shift in the narrative towards "building." As someone who self-identifies as a hopeless techoptimist, I'm deeply aligned with the desire to build and to innovate.

For some reason, when I came across this post titled "it's time to publish build", or rather, the prevalent sentiment it represents in the current cultural cycle of technology, I couldn’t help but think of the above confessionary note from that anonymous Chinese VC.

I grew up in a Metropolitan city in China, and my childhood apartment was located in the central business district, where some of the most innovative businesses set up their headquarters or stores. During the Chinese consumer Internet bubbles, rows of yellow scooters and bikes lined up under the old pine trees, and dozens of ride-share companies with their multi-color cars raced through the streets.

There were automated gyms, manless supermarkets, and robotic package pick-up and delivery drop-off kiosks. All of these, in the past years, have now been replaced by floor-to-ceiling glass-paneled EV stores reminiscent of an Apple store. They're all beautiful, gigantic, and new. On one street, we can find four various luxury EV brands with their showrooms.

The sensation is surreal: an urban landscape that is neither post-apocalyptic nor utopian, but something in between. Gleaming towers and elegant showrooms, yet overlaid on the fading scaffolding of yesterday’s hype cycles. The visible churn of capital, ambition, and state planning has delivered a kind of surface modernity, but left a deeper ambiguity about what is being built and for whom. Industrial spectacles have upstaged the consumer internet’s improvisational energy.

In China’s innovation economy, the evidence of progress is everywhere, but the oxygen for genuine experimentation feels scarce. We see the stifling consequences mentioned in that confession:

• Investment now deprioritizes analytical rigor and intellectual arbitrage.

• Overemphasis on “hard tech” disregards genuine commercial viability.

• Asset-heavy investments inflated by state demands unnecessarily burden enterprises.

• Investors become mere matchmakers, diluting investment integrity.
  Short-term profit incentives foster inconsistency and ethical erosion, hindering sustainable growth.

What’s really at stake is not some tired “hard tech versus soft tech” or “state versus market” dichotomy, but the way both systems turn collective delusion into fuel, and how the shape of the delusion ultimately determines whether anything real and enduring gets built.

It’s easy to reach for common sense, like decentralized systems are messy but resilient, while centralized ones are sleek but shatter under real stress. But what’s genuinely striking is how, for all their differences, both systems tend to converge in the end, each constructing elaborate performances to sustain collective belief, each inadvertently sanding down the friction that makes actual discovery possible. The symptoms may be presented differently, but the underlying dynamics appear eerily similar. The pathologies rhyme even if the costumes are different.

Recently, I had a conversation with the wonderful Linus Lee about the dual capacity of markets and states: each can create collective delusions, shaping what people believe is possible, valuable, or “real.” He called it “top-down vs. bottom-up delusion,” which I found fitting.

States and markets both create reality distortion fields. Each can whip up their flavor of mass consensus, printing stories that nudge people to believe, build, and occasionally, break the rules of what’s possible. But the shape of the spectacle, and how it collapses or mutates under pressure, remains distinct.

I start to think the greatest threat to innovation isn’t picking the wrong system (state-driven vs. market-driven) but optimizing any system for innovation’s performance rather than its messy, uncertain, and necessary reality.

When innovation becomes purely a spectacle, whether performed for party officials in Beijing or venture capitalists in Silicon Valley, we lose the very conditions that make breakthroughs possible. We trade the messy, uncertain, sometimes embarrassing process of genuine discovery for the clean, certain, and ultimately hollow choreography of progress.

In China and many parts of the world, it’s a top-down spectacle: reality is shaped by quotas, policies, or directives from a ministry or a mayor. There’s a kind of surgical coordination when HQ says “build hard tech,” whole industries reorient, capital sluices into “priority sectors.” Careers rise or fall on how precisely you color inside the prescribed lines. It’s beautiful in its clarity, and terrifying in its inertia. Progress can be forced, but so can fictions. Delusions persist because they’re expensive to puncture. Nobody wants to be the person who says the numbers don’t add up until the reckoning, when, all at once, they don’t.

Meanwhile, markets specialize in bottom-up spectacle. Delusion bubbles up in infinite little pockets—crypto Discords, startup accelerators, Twitter threads that spiral into new movements, garage-scale “consensus hallucinations” that occasionally crystallize into something real. The system corrects: ideas die, memes lose steam, funding in specific sectors evaporates with no sight of liquidity, and the players who bet wrong get reset.

Both top-down and bottom-up systems engineer collective delusions, each propagating myths, whether of omniscient planners or visionary founders. They differ not in creating illusion but in how swiftly and transparently illusions collapse.

The myth of the lone founder is as much a spell as the myth of the central planner. Any nascent sector—city, startup, policy regime—only gets off the ground because enough people are willing to lie a little to themselves and to others, or to believe the impossible, until the prototypes, products, or movements grow sturdy enough to survive contact with reality.

But the scale and repairability of these collective delusions couldn’t be more different. In markets, feedback is relentless and public. Anyone can opt out, money dries up, and the scoreboard is visible to all. You can run dozens of experiments, kill your darling, burn the bridge, start from scratch, and there’s always another lane to merge into.

Market delusions burn out quickly but can be wasteful and chaotic. Top-down systems can sustain focus on long-term goals, but they risk defending failing strategies for too long. Each approach trades different types of resilience against different vulnerabilities.

Top-down, imposed delusions have their own metabolism. Feedback can be sticky and slow, with dissenters shuttled into silence or the truth warehoused “for later.” Systems like these can defend a failing script for years, sometimes decades, but they are not always rigid or inert. The Chinese state, for example, has demonstrated the ability to iterate both policy and industry direction at a breakneck pace. Mandates can shift overnight; priorities are reshuffled according to new geopolitical realities, financial tides, or technological breakthroughs.

Yet flexibility has costs: bureaucratic feedback loops and suppressed dissent mean updated scripts rarely confront underlying truths. Even as new directives and strategies replace the old, a deep inertia can persist: failures and frictions get papered over, and self-preservation takes precedence over uncomfortable recalibration.

So when top-down delusions finally break, the magnitude is often systemic: whole industries, generations, or social contracts left to rot or collapse quietly, sometimes all at once, sometimes in swift, coordinated shudders. The real danger isn’t just stasis, but the way flexibility without genuine bottom-up challenge can mask how long the system has been running on theater rather than truth.

Most critically, the flexibility and “escape hatches” are fundamentally different. A system shaped by markets encourages a hundred parallel stories, even if most wilt. In top-down regimes, there’s only one reality, until suddenly, there isn’t. When the script fails, it fails.

So what matters isn’t just the presence of delusion, but the architecture of error recovery: how quickly can the narrative evolve when the script gets out of sync? Who gets to call the bluff and write the next act? Where is dissent metabolized and made useful, and where is it suffocated into compliance?

This is why discerning the scale and metabolism of a system’s delusion matters. When the spell breaks, robust ecosystems allow pathways for dissent and script-flipping, spaces where truth can be metabolized and new realities quickly emerge. Fragile ones get stuck in denial or scramble for hasty rewrites. Resilience isn’t about preventing delusion, but about how rapidly and transparently a system can surface counter-narratives and rebuild from what’s left, turning collapse into renewal rather than stasis or denial.

The best innovation systems foster productive disillusionment: they create conditions under which delusions inspire action, yet swiftly surface contradictions through open, continuous, and rigorous feedback. Healthy ecosystems cultivate productive skepticism, inviting regular resets without shame or catastrophic consequences.

The danger for the West isn't just mimicking China's state-directed approach; it's falling into the false binary of "PDFs versus building." Recent calls to redirect university funding toward trade schools and manufacturing may overlook the fact that the problem isn't that we've too much theoretical work, but that we've lost the conditions for genuine discovery in both theory and practice.

Yes, academic institutions are no longer the origins of the best ideas. Still, what the "PDFs" symbolize, the ability and freedom to propose bold ideas, followed by the will and agency to validate such ideas, still makes innovation possible. The Bitcoin whitepaper is an example, as is the "Attention is all you need" paper. Even a blog post like "Why software is eating the world" has been an empowering call to action for all sectors, not just technology, to reexamine how things can be done and how they can be done better.

The U.S. could maintain such a leading position precisely because of the built-in systematic inefficiency. World-changing breakthroughs and innovations, by definition, cannot follow any predictable patterns and can emerge from anywhere. The real lesson from both the Chinese VC's confession and America's current "building" moment is that breakthrough innovation requires defending the weird, the ambiguous, the half-finished spaces where genuine surprise can emerge.

To build is as much an act of courage as an act of precision. The danger is not in picking the "wrong" model, but in becoming so fixated on signaling compliance, whether with the state, with markets, or with metrics, that we extinguish the subtle, irreducible weirdness from which real progress comes.

The real choice isn’t between "building" and "not building," but between fostering systems that reward actual breakthroughs versus those that merely celebrate their appearance. True innovation thrives in cognitive friction; spectacle eliminates it.

Innovation isn't choreography or theater; it’s dangerous, uncertain, and often irrational, exactly why genuine breakthroughs surprise us. Sometimes, it’s a PDF read by three people that remakes the world. Sometimes, it’s a factory line no one believed in, until it worked.

The only reliable truth: breakthroughs are, by definition, unpredictable. They require room to breathe and room to fail.

Further readings

• What Goes Wrong by Jessica Livingston

• The Promise and Pitfalls of Government Guidance Funds in China

• The Role of Government Guidance Funds in Financing Innovation in China

This is the blog post version of the paper we published with expansion on more actionable tactics. You can view the full paper here.

Subscribe now

Rewind to an idea that once lit up machine learning: "Attention is all you need," the breakthrough that showed AI models could focus on relevant parts of data without complex memory systems.

Yet, the next decade may prove that attention is only as powerful as the context you feed it. The race for AI dominance has shifted from who has the best models to who controls and understands user context.

Startups like Limitless are shipping $99 pendants that clip to your shirt and record every conversation, promising to transcribe meetings and answer questions based on your accumulated interactions. Similar devices like Friend's are betting on much more personalized companions that can provide just-in-time personal interactions based on any situations. The race for context is underway.

This essay distills key insights from our recent collaborative paper into actionable tactics for product builders and designers.

The Promise of (and the Race for) Context

A few definitions to get on the same page:

Context: All signals available to a product, including environmental cues, device state, user history, and intent.

Context Window: The quantity and recency of information processed by a model.

Memory: Persistent context carried forward across sessions.

Personal Context Infrastructure: Systems that aggregate, store, update, and share context between services.

Context Boundary: Rules defining what information is accessible and how memory is managed.

​​Effective context-aware systems operate across four layers:

• Immediate: Current conversation or task.

• Session: Today's work or project.

• Personal: Preferences, patterns, relationships.

• Environmental: Location, schedule, available devices.

For as long as we’ve dreamed of AI, we’ve longed for machines that not only “get” our explicit asks but seem to understand us, the way a longtime friend finishes your sentence or a creative partner anticipates your next thought.

We want AI to evolve from a mere notetaker to a genuine collaborator, a presence that knows when to nudge, when to listen, and when to step back.

The most valuable resource isn’t algorithms, compute, or even just data. It’s context: the mosaic of real-time signals, history, recent mood shifts, stray calendar invites, and the live-wire pattern that is “you, right now.”

There’s a collective land grab for context. Who gets to see it? Who stores it? Who profits from it? Whoever answers these questions wins the right to shape what AI is for the next generation. Business models are already emerging around context:

• OpenAI's GPTs: User generates context and memory by using the product, OpenAI hosts and monetizes

• Notion's AI: Context stays in user workspace, subscription model

• Google's Ultra AI: Free context analysis across product surface area that can subsidize ad targeting

• Apple Intelligence: On-device context processing as premium hardware differentiator

From Data Hoarding to Context Awareness

Historically, platforms vacuumed clicks and scrolls into massive datasets. Today, with smarter models and expressive interfaces, AI builds "digital twins," live models of user intent and state. Recent context now vastly outweighs stale historical data.

For example, a health-tracking wearable that alerts you based on your current heart rate, hydration, or stress levels rather than past averages. Smart home devices that adjust lighting, temperature, or music based on immediate activities or conversations, not historical usage patterns. Productivity apps like Linear or Notion highlighting tasks that are actively being discussed in real-time collaboration.

TACTIC #1: Prioritize Real-Time, Portable Context

• Focus on real-time signals over legacy data, like live user interactions, real-time sensor inputs, or immediate environmental conditions.

• Design context ingestion pipelines optimized for immediacy, such as instant transcriptions or live sentiment analysis.

• Make "context state" objects explicit and actionable, like real-time dashboards showing current user states or live context feeds for immediate decision-making.

Context Is Generated, And Earned

Context isn’t something you can just scrape or surveil. You have to earn it. The interfaces we choose—whether chat, camera, calendar, or code editor—are invitations. They shape not only how much context we’ll willingly share, but also the kind that gets generated. A design tool will surface creative patterning. A health wearable will log your physical rhythms. An intimate AI chat will prompt you to disclose the messy, heartfelt stuff you’d never type into a search box.

The big unlock will come from combining data from different domains. Your Spotify listens, Dexcom data, and Notion scratchpads are all different “eyes and ears” on your life. The winner might just be whoever stitches the most relevant slices together in a way you actually trust.

TACTIC #2: Enable Composable, Multi-Source Context

• Make it seamless for users to permission and de-permission slices of their context (per product, per task, per agent).

• Build apps/agents/APIs/protocols that let users stitch context across domains (calendar + health + writing tools + chat).

• Don’t fight context silos. Design for “cross-context” products that combine the few signals that matter for each job.

The Context Stack

Success comes from mastering three technical layers working together.

• Collection Layer: APIs, sensors, and behavioral signals. GitHub Copilot doesn't just read your code—it notices your coding patterns, preferred libraries, and when you're most likely to introduce bugs.

• Processing Layer: Where raw signals become intelligence. This handles context synthesis (connecting dots across data sources), privacy filtering (what gets remembered vs. discarded), and relevance ranking (what matters right now).

• Application Layer: Context becomes action. The AI that finishes your sentences, the assistant that avoids scheduling calls during your kid's soccer practice, the design tool that suggests your recent color themes.

The winners either build all three layers (OpenAI's full-stack approach) or dominate one layer while integrating seamlessly with others (Calendly owns scheduling context, Linear owns dev project context).

This architecture explains why context portability is hard. You're not just moving data files, you're losing the processing logic that learned your patterns. Trust becomes a competitive advantage: users share more context with systems they trust, creating a flywheel effect across all three layers.

Examples of Context Stacks in Applications

Cursor: coding agent

• Precise control of intent and state context.

• Uses @-symbol syntax for specific context targeting.

• Minimalist approach, prioritizing relevance.

Cursor proves that context minimalism beats maximalism. "Too much irrelevant context dilutes the signal." Success comes from giving developers precise control over what matters when.

Spotify: personal DJ

Spotify's AI DJ demonstrates how layering multiple context signals creates deeply personalized experiences:

• Combines immediate, session, personal, and environmental contexts.

• Invisible context collection through seamless interactions.

• Real-time narratives for personalized user experience.

Spotify succeeded by making context collection invisible while delivering immediate value. Users don't fill out preference forms, they just listen while the system learns.

Personal Context Servers

Several of us in the group played with the idea of a local, personal “MCP” (Model Context Protocol). Tina previously wrote about it on People as Files. Imagine every insight about you, from personality type to skill profile to emotional pulse, is stored in a sandbox you control.

The dream: a portable context engine that travels with you, layerable across services, encrypted and privacy-respecting by default. In theory, you could “train” a new chatbot on your real personality—or let a digital agent represent you—without starting from scratch every time you switch apps.

In practice, privacy and security challenges remain complex. But the direction is clear: a future where your context truly belongs to you, freely movable with permissions tailored to your preferences.

TACTIC #3: Move Toward Portable Context “Profiles”

• Let users export/import raw context and trained “persona” objects (as files, APIs, or via open protocols).

• Make it explicit what part of the user’s context your product ingests, and let them audit/edit the feed.

• Expect the emergence of “context wallets,” products that manage, update, and route slices of user context to downstream agents.

Why Context Stays Locked Up

Legacy platforms have no reason to help you escape. The business model is simple: lock in as much context as possible, then charge rent (in money, attention, or both) for access to your own life. Regulation (GDPR, DMA) is starting to force cracks in these walls. The friction is part of the moat.

​​But the challenges run deeper than business incentives. Previous attempts at portable identity, from OpenID to decentralized social protocols, failed not just due to platform resistance, but because context interoperability is genuinely hard. Context isn't just data; it's data plus the processing logic that interprets it. When you export your Spotify listening history, you lose the algorithm that understands what "listening to Radiohead at 2 AM" means for your emotional state. Technical standards can share the what, but struggle with the why and how.

Context exhibits powerful network effects that work against portability. Each additional piece of context about a user doesn't just add linear value,it creates exponential improvements in personalization. A company with 80% of your context can provide dramatically better service than one with 20%, making it economically irrational for users to fragment their context across providers. This creates a natural monopolization force where winner-takes-most dynamics emerge.

The computational costs also become staggering. Maintaining a rich, multi-layered context for millions of users requires massive infrastructure. Every additional context signal exponentially increases processing complexity. This creates a natural centralization force, only tech giants can afford to maintain context at scale, reinforcing the very lock-in we're trying to escape.

Most critically, we face the privacy paradox: users simultaneously crave deep personalization ("know me like a best friend") while fearing surveillance ("but don't watch me"). This isn't irrational, it reflects the genuine risk that context powerful enough to anticipate your needs is also powerful enough to manipulate them. The same context that helps an AI therapist provide better support could be weaponized for targeted exploitation. No technical architecture fully resolves this tension.

So what breaks the cycle? Maybe it's regulation, maybe it's collective consumer demand, or maybe it's "killer apps" that make it genuinely worth your time to aggregate your context somewhere new. If a product or protocol helps you get real value such as better recommendations, more meaningful decision support, even a sense of feeling "seen," you'll move. If not, inertia wins and the silos persist.

TACTIC #4: Design for Context Leverage, not Lock-in

• Empower users through aggregated, portable context.

• Build context-aware features that provide immediate value.

• Be the product users actively choose to bring their context into.

• Be the “killer use case” that gets users to bring their context in.

Trust, Intimacy, and Boundaries

People are already more candid with AI than with most humans. When your AI agent “gets” you, it can feel downright magical. But the line is thin. Premonition, overdone, is unsettling or even creepy. Worse, breaches of context trust feel like personal betrayals, whether it’s a chatbot forgetting your favorite color or a search history resurfacing at the wrong moment.

AI systems need to support radically different context preferences across domains. Some users want everything unified, others want sharp boundaries between work and life, or between public and private selves. Agency isn’t just about context control; it’s about the ability to curate which “self” you bring to each interaction, and to see (and edit) what your agents see.

TACTIC #5: Build Context Boundaries and User Controls

• Let users toggle between “public,” “work,” “private,” or custom context scopes for different agents.

• Expose “what your agent sees” as an auditable, interactive surface (not a black box).

• When in doubt, surface ambiguity: ask if the agent should remember/discard/refine certain facts.

From Prediction to Action

The true potential of context-aware AI lies not just in predicting but in autonomously acting on user intentions. Not just “Do you need new vitamins?” but “Should I order them?” All of this depends, again, on context, knowing enough to take action, but not so much as to overstep.

The interfaces will feel more ambient, more “always on,” more gently anticipatory than the rigid forms and feeds of the web 2 era. And as these agentic systems mature, we’ll see both specialized and general-purpose versions, from the domain-specific assistant that reigns in health or finance to the “big brain” generalist that orchestrates across your life.

TACTIC #6: Productize Context-to-Action Loops

• Enable one-click or agentic follow-through (“Do this on my behalf”).

• Allow users to preview, veto, or modify actions based on current context state.

• Instrument the “last-mile” of automation: what context drove which actions, and with what confidence.

End State: Agency or Enclosure?

We're at an inflection point. The AI race is shifting from model performance to context mastery. The winners won't be those with the biggest models, but those who build the most trusted context infrastructure.

• If you're a builder: Start with one killer use case that makes context portability irresistible. Focus on immediate value, not abstract promises. Remember: Spotify didn't pitch "context awareness"—they just made music feel personal.

• If you're a platform: Choose your strategy now. Will you lock in context and milk it or become the trusted context layer others build on (the Stripe playbook)?

• If you're a user: Your context is your leverage. Support products that give you control, not those that mine you for data.

The technical challenges are real. Interoperability is hard, computation is expensive, and privacy is paradoxical. But these are engineering problems, not laws of physics. The first products to crack the code will trigger an avalanche. Once users experience a truly portable, powerful context, there's no going back to silos.

Context is all you need, but only if you build for user agency, not addiction; for portability, not lock-in; for trust, not surveillance.

If these themes resonate, or if you want to dig into the full slightly more technical and design detail, you can check out the paper here.

“Everywhere, everything is ordered to stand by, to be immediately at hand, indeed to stand there just so that it may be on call for further ordering.”

— Martin Heidegger, The Question Concerning Technology

Earlier this week, I shared a screenshot of ChatGPT 4o’s thoughts on life in NYC, what makes NYC so wonderful. The response felt so specific and human that I had to post it on X.

It spread far wider than I thought, and I was equally surprised by the reception of a rather frivolous prompt. While most people found the AI’s answers uncannily relatable, there were also waves of revulsion and disbelief. To the latter, the exchange seemed to embody everything hollow and culturally tone-deaf about a world that they believe has been defiled by tech.

But what exactly was so provocative that it created such visceral backlash (and its resonance)? It’s not just that such flowery prose can be created by AI; perhaps it’s the feeling that machine-generated content can mimic the mechanical shape of emotional expression, while lacking it completely.

Unlike automation in spreadsheets, code reviews, or grading homework, which we view as mechanical and unoriginal drudgery, AI-generated prose attempts to inhabit domains we instinctively associate with intimate subjectivity: observation, memory, yearning, regret, which are unique to us. When an LLM “describes the rain” or tries to evoke loneliness at a traffic light, it produces language that looks like the real thing but does not originate from lived experience, and as a result, feels almost intrusive. You didn’t even experience this, how would you know?! Is the AI’s view of New York a real cultural artifact, or just a reflection of our data? The closer it gets to real expression, the more unsettling its lack of personal context becomes.

Human communication carries the implicit promise that someone who’s capable of registering what we mean is on the other side. Sherry Turkle’s work on simulation shows how digital environments can signal understanding without the inner life those signs can indicate. We’re drawn in by the simulation because it mimics the surface of a relationship so well, as evidenced by the rise of the like Character AI.

Philosopher Byung-Chul Han pushes in a complementary direction: a culture of total transparency and frictionless legibility strips away the opacity and interiority that mark a genuine other. When everything is immediately readable and endlessly available, the depth that makes another mind worth recognizing flattens.

“Today’s narcissistic “achievement-subject” seeks out success above all. Finding success validates the One through the Other. Thereby, the Other is robbed of otherness and degrades into a mirror of the One, a mirror affirming the latter’s image.”

What’s a more perfect mirror than the LLMs? Why do we fear what we see in the mirror?

The Nature of our Discomforts

To grasp this discomfort, we must look at what sets AI apart from older technologies. The heart of the matter is not that “AI is evil” in the classic sense of malice or cruelty. In fact, it is the opposite. What truly unsettles us is AI’s chilling indifference, often called “slop,” a blankness that feels like the absence of a soul.

Software doesn’t hate, plot, or hold a grudge. It optimizes. That optimization includes efficiency in language, created stochastically to fulfill a certain objective, rather than from a place of meticulous intent. This is where Hannah Arendt’s insight on the “banality of evil” feels uncannily relevant. In her words:

“The essence of totalitarian government, and perhaps the nature of every bureaucracy, is to make functionaries and mere cogs in the administrative machine out of men, and thus to dehumanize them.”

“The sad truth is that most evil is done by people who never make up their minds to be good or evil.”

(Eichmann in Jerusalem)

Arendt warned that true horror often arrives not at the hands of monsters, but through “nobodies” who “never realized what they were doing,” mindlessly turning the crank. It is not the tyrant’s bloody grip that propels the machine, but the process itself, moving forward without intention. “In place of intention, we get process.” The algorithm does not plot; it simply runs.

We crave stories, searching for intent behind the curtain—heroes to cheer, villains to blame. Yet in the cathedral of the algorithm, we find only emptiness: “There is no one there.” That void suffocates us more than any scheming demon could. We want our villains to act with purpose; instead, we stand bewildered before an indifferent god.

Lila Shroff’s luminous speculative fiction foresaw, with uncanny clarity, how squeezing human communication for efficiency can breed tragedy. In her story, college friends Kate and Hannah drift apart after an AI summary app in Hannah’s inbox misses the subtle signs of Kate’s emotional distress and depression.

Our minds, shaped by tribal politics and campfire tales, are unprepared for the moral fog of a system with no soul to save or condemn.

Our discomfort with this indifference manifests in a predictable pattern: we try to turn AI into a comprehensible villain. In reference to Girard, every society is built, in part, on the ritual of scapegoating, a mechanism by which collective anxieties, rivalries, and fears are projected onto a villain or outcast, thereby restoring temporary order through exclusion or blame. We yearn for something or someone to hold accountable, a focal point for our moral clarity and rage. Today’s technology is a perfect canvas for those projections.

The terror is existential. AI shakes the stories we tell about our uniqueness—our creative spark and our sense of agency. Western moral philosophy, from Kant to Nussbaum, revolves around treating humans as ends in themselves, never just means. To reduce a person to a function isn’t just a philosophical error, it cuts at the heart of our dignity.

And yet, the logic of AI, especially as business infrastructure, makes this reduction feel inevitable. Algorithms sort us by engagement scores, by revenue-per-user, by “propensity to churn.” The most controversial AI applications—social credit systems, predictive policing, “optimized” hiring—aren’t dystopian because they are cruel, but because they are perfectly indifferent.

What if this tendency to villainize AI obscures a deeper truth about how technology reshapes not just our tools, but our entire way of seeing the world?

Heidegger’s Warning: The Danger of In-Framing

Long before AI, philosopher Martin Heidegger anticipated this transformation.

In The Question Concerning Technology, Martin Heidegger warned that technology is not just a collection of tools, but a way of seeing the world, a revealing that both illuminates and conceals. Heidegger’s “enframing” (Gestell) describes technology’s insidious power to frame everything (nature, humans, even time) as “standing reserves,” ready to be ordered, manipulated, consumed.

“Everywhere everything is ordered to stand by, to be immediately at hand, indeed to stand there just so that it may be on call for further ordering.”

In this view, AI is not merely software deployed for business efficiency. It is a force that recasts reality itself through the lens of optimization, availability, and control. The data labelers, the Uber driver, and even the AI-augmented knowledge worker, anyone below the LLM all at some point become “resources” in a perpetual state of readiness.

But Heidegger wasn’t a Luddite. He saw a paradox: technology both reveals and conceals. It produces astonishing new worlds, but only by flattening the world into what can be stored, indexed, and summoned at will. In the digital age, this is literal. Every social act, every click or hesitation, becomes a data point, a potential input to someone’s model.

In the workplace, to be called a “machine” is, weirdly, a compliment: a tribute to one’s measurable output, reliability, and consistency. Productivity software rewards us for becoming more like the systems we build: efficient, predictable, and always on.

But as soon as AI demonstrates even a flicker of poetic intent, a gesture toward the unquantifiable—art, longing, vulnerability—many repulse. Some do so because they sense (correctly) that computers are still not good at this; others, because they sense that it is only a matter of time. Both reactions betray a fear that the territory of the soul, of human meaning, is shrinking. The room for contemplation, leisure, and error is increasingly defined by what resists digitization, until, suddenly, even error is modeled.

AI systems present themselves as agents of transparency, optimization, ranking, and matching with a supposed objectivity. But their inner workings, from transformer weights to diffusion mechanisms, are often profoundly opaque. Even the researchers behind the models struggle to explain why certain outputs appear, why some biases manifest, or how adversarial data alters meaning. This opacity is not just technical, but philosophical, a concealment at the core of the digital order.

Efforts at “interpretable AI,” “explainability,” or algorithmic audits echo Heidegger’s call toward aletheia, the ancient Greek notion of unconcealment, or truth-as-revealing. The contemporary push for AI transparency is, at its best, an effort to reclaim dignity and agency within systems that profit by rendering us invisible to ourselves.

The contemporary push for AI transparency is, at its best, a bid to reclaim dignity and agency within systems that profit by making us invisible to ourselves. But even this may not be enough to address the existential challenge AI poses.

The Evangelion Vision: Human Instrumentality

"Any where can be paradise, as long as you have the will to live, after all you are alive, you will always have the chance to be happy. As long as the sun, the moon, and the Earth exist, everything will be alright."

— Yui Ikari from Evangelion

In Neon Genesis Evangelion, the “Human Instrumentality Project” offers to dissolve all suffering through perfect togetherness. No more pain, no more loneliness, every fragment of individual consciousness merged into a single, undivided mind. It is paradise, and it is annihilation.

Evangelion’s answer is ambiguous. The protagonist, Shinji, and his shattered friends are offered relief from their existential ache, but at the price of agency, individuation, and the possibility of meaning. The show is a fever dream of depressive suffering and the refusal to submit to a world without boundaries, without the dignity of suffering one’s own fate.

What makes us human is not the absence of pain, but the experience of it as ours. AI, in its idealized form, offers collective agency, a solution to loneliness, an optimization of every choice. But as both Heidegger and Evangelion warn, the cost may be unbearable: the flattening of selves into a seamless, computation-ready mesh.

Evangelion's ambiguous ending, neither full acceptance nor total rejection of instrumentality, points us toward a more nuanced response to our technological moment.

Heidegger’s Saving Power

This brings us back to Heidegger, who offers a paradoxical hope.

Yet, as Heidegger’s essay also insists, the very force that threatens us may harbor its own “saving power.” Recognition of technology’s danger is itself a call to awaken.

But he doesn’t ask us to simply retreat, become luddites, or declare defeat in the face of technocratic sprawl. Instead, Heidegger compels us to do something much harder: to see the world as it is being reframed by technology, and then to consciously reclaim or reweave the strands of meaning that risk being flattened.

The “saving power” arrives not as a counterweight, but as a paradox: we are awakened to the danger precisely through contact with it. The same algorithmic indifference that unsettles us may also jolt us into a higher vigilance, a refusal to hand over the entirety of our experience to optimization, market logic, or digital control. The very anxiety these systems produce is a clue: something vital, unquantifiable, and irreducibly human still resists.

This isn’t about throwing away the tools, but about wrestling them into alignment with what we find sacred or essential. We won’t find salvation by escaping technology, but by using our awareness—our capacity for critique, ritual, invention, and refusal—to carve out room for messiness, for mourning, for risk, and for deep attention. The saving power is the act of remembering, in the heat of technical progress, to ask: “What space remains for meaning? For art? For wildness, chance, suffering, and genuine encounter?

AI is not inevitable fate. It is an invitation to wake up. The work is to keep dragging what is singular, poetic, and profoundly alive back into focus, despite all pressures to automate it away.

Coda: A Way Forward

What unsettles us about AI is not malice, but the vacuum where intention should be. When it tries to write poetry or mimic human tenderness, our collective recoil is less about revulsion and more a last stand, staking a claim on experience, contradiction, and ache as non-negotiably ours.

But perhaps that defensiveness, too, is a signpost. What if the dignity of being human is not to stand forever outside the machine, but to insist, even as the boundaries blur, on those things that remain untranslatable—mess, longing, grief, awe, strangeness? What if technocracy and romantic withdrawal are not our only options, but two poles of a far richer field, a frontier where we move, trade, shudder, and insist on meaning?

Call it “the saving power,” call it “instrumentality,” call it the refusal to become seamless. Our task is not to panic when the machine gets close, nor to mythologize our differences into oblivion, but to take seriously the ongoing work of being human: suffering, loving, resisting reduction, and making art out of what refuses to compute.

That, for now, is enough. Maybe the challenge is to keep returning to these questions—not to solve them, but to stay alive inside them.

Further Reading & References

• Heidegger, Martin. “The Question Concerning Technology”

• Arendt, Hannah. “Eichmann in Jerusalem: A Report on the Banality of Evil”

• Nussbaum, Martha. “The Fragility of Goodness”

• Neon Genesis Evangelion (Anime), Hideaki Anno, 1995

• On interpretability and AI transparency: Doshi-Velez & Kim, “Towards a Rigorous Science of Interpretable Machine Learning”

• On the loss of dignity: Michael Sandel, “What Money Can’t Buy”

• On the human instrumentality problem: Robert Nozick, “The Experience Machine”

This special edition is a collaboration with my friend Yondon Fu. We bonded instantly over our compulsive drive to tinker at the edge—most recently, diving into new consumer frontiers unlocked by AI. Yondon brings deep technical expertise from his experience as co-founder and Head of Engineering at Livepeer, where he architected decentralized media protocols that power millions of video streams daily. Together, we've been imagining the transformative experiences possible when cutting-edge AI intersects with one of our favorite media, gaming.

Gaming began as pixelated wish-fulfillment, limited by hardware yet boundless in imagination. Each technological "no" was met with an artist’s defiant "yes, but what if?" Now, at the threshold of AI-native worlds, we’re about to transcend those limits entirely, venturing into infinite playgrounds that evolve as swiftly as our dreams.

Fast forward to today, games have exploded into boundless forms: part theater, part recursive universe, part psychological mirror. AAA studios sculpt worlds the size of continents; indie creators conjure myths from dorm rooms.

As avid gamers, we find ourselves dreaming of that elusive “perfect game” once again, now tantalizingly within reach through AI. They are ever more expansive, personalized, and generative. And it won’t merely be a fresh coat of paint or an incremental upgrade to existing engines, but rather reweave the very fabric of creation and reshape the contours of experience, destabilizing familiar roles for both developers and players alike. What is left to be redefined are storytelling, agency, and imagination itself, boldly testing and transcending the limits of our current infrastructure.

What Does an AI-Enabled Game Look Like Today, and Tomorrow?

Today, AI-enabled gaming teeters intriguingly between experimental promise and practical reality. Platforms like Rosebud.ai showcase how mini-games can be rapidly crafted with AI’s assistance, resulting in low-cost productions that, while high in creative variance, typically face steep user retention challenges and distinct power-law dynamics.

On the other side of the spectrum are games crafted meticulously using traditional tools like Blender or Unreal Engine, which embody AAA production readiness, are richly immersive, yet demand significant expertise and technical investments.

The true creative frontier lies in the liminal space between, marrying the speed and flexibility of generative production with experiences people want to revisit, linger in, and hold close. This middle ground is misty, unmapped, an expanse of questions and untapped forms.

Here are some possibilities:

AI-crafted narrative worlds that evolve endlessly.

Imagine vast, intricate universes where storylines dynamically shift based on the collective decisions and individual choices of millions of players. AI can analyze player interactions and craft complex narrative pathways in real-time, making every journey uniquely personal yet interconnected. These worlds would continuously evolve without human scripting, offering endless replayability and deep narrative immersion that responds fluidly to emergent player-driven scenarios.

In an AI-native MMO inspired by Dwarf Fortress, every player's action, from planting a tree to starting a war, subtly alters the world’s history. The AI not only generates lore but also weaves ongoing geopolitical drama, emergent factions, and historical “what-ifs.”

Such systems rely on multi-agent world simulation, using LLMs for dynamic dialogue, adaptive NPC behaviors, and track social and economic relationships. A persistent narrative engine digests logs, infers meaning, and continually seeds future conflicts and mysteries, surfacing rich, unexpected twists that even devs didn’t directly design.

Gameplay that morphs to you in real time

Unlike ever-expanding AI-generated worlds, adaptive gameplay loops are about moment-to-moment change: the core mechanics, challenges, and pacing shift in direct response to your unique style and needs. Here, the AI becomes an omnipresent “game master,” learning from your decisions and tuning the rules of play to keep you on the knife’s edge of engagement.

If you solve challenges effortlessly, the game senses complacency and amps up complexity on the fly. If you’re struggling, the AI discreetly lowers the difficulty curve, offers context-sensitive hints, or even reconfigures level layouts to ease frustration. The gameplay experience as a result should feel “just challenging enough,” neither too punishing enough to keep the player in flow state. Unless, the difficulty of the game alone is a feature (looking at all the tortured souls of Souls players).

Edge models can allow instant adaptation without server lag, while meta-learning lets the system generalize “player archetypes” and share insights across the user base. Advanced games may use embedded A/B testing to prototype new loops on the fly.

Richly personalized NPCs and emergent storylines

In an AI-native RPG, an NPC you helped in chapter 1 remembers a song you liked and, 50+ hours later, references it poignantly as you face a major choice. Another NPC, trained on subtle sentiment analysis on your playstyle, mirrors your anger or hope, sometimes even “misunderstanding” or developing a grudge based on your past dialogue tone. These bonds become as real as friendships, with evolving arcs that span dozens or hundreds of hours.

A player can write: “I want to rescue a dragon from a haunted library using nonviolent negotiation.” Seconds later, the game responds with a play-balanced, coherent quest that is threaded seamlessly into the world. If abusive or exploitative content is detected, the AI's real-time filters rebalance or reject it, ensuring fairness.

The design pattern relies on persistent memory graphs attached to each major player-NPC interaction, rooted in embeddings of conversations, emotional signals (voice/text), and external context (game events, time played). Fine-tuned transformers allow “memory retrieval,” while generative models use that context to craft contextually relevant dialogue or events. Some studios experiment with hybrid on-device/edge cloud inference to keep memory costs manageable and secure.

Evolution of AI as A Native Medium

V1 (Text + Images): Where we are today

AI companions and role-playing applications like CharacterAI and SillyTavern leverage text and image synthesis, delivering intimate, narrative-driven experiences that serve as foundational stepping stones.

V2 (Games): The next horizon

In the next-gen Sims or Inzoi, each simulated neighbor isn’t just a scripted automaton but a mini-agent with their own evolving biography, values, and mood, utilizing generative social modeling. If you play as a reclusive artist, NPCs may proactively invite you to events or create rumors, while your AI dog’s personality adapts to how you treat them.

These games can leverage probabilistic agent-based models, hierarchical planning, and context-aware LLMs, where each character retains their own memory buffer and “emotional vector.” High-fidelity social simulation requires not just procedural animation, but emergent schedules, gossip transmission, and dynamic social norm learning, a fusion of “The Sims” mechanics with multi-modal AI.

V3 (“Worlds”): Emerging universes

Creations from teams like Supersplat, DeepMind, OdysseyML, and World Labs represent entire universes conjured, managed, and evolved by AI. These worlds are persistent, dynamic, and continually responsive to collective player actions, growing organically based on aggregated player behaviors, preferences, and interactions, creating vast, living ecosystems rich in emergent complexity.

Why Now for AI-Native Gaming?

• Unprecedented creative scale: AI-native games allow for near-infinite content generation. Where classic MMOs recycle dungeons, AI games can generate new worlds, enemies, and plots every hour—customized by individual or collective desire. This unlocks replayability and a “living history” impossible with static development pipelines.

• Player agency and identity: By deeply personalizing experiences and NPCs, AI-native games blur the line between authored and emergent narrative, giving players a sense of agency previously reserved for tabletop RPGs and collaborative fiction. Identity becomes entwined with the world’s evolution, changing how we understand both “winning” and “belonging.”

• Democratized development: The barrier to entry for creating sophisticated games drops dramatically as AI handles the heavy lifting of content generation, QA, and even live operations. Indie designers, solo creators, and even players themselves can sculpt worlds and stories with less friction and at lower cost.

• New economic models: AI games introduce new incentives. Players can earn, co-create, and co-govern digital spaces. Reputation, digital goods, and even AI-authored myths become tradeable and meaningful, enabling decentralized economies and novel business models.

• Technical feasibility and momentum: With the explosion in multimodal models, cost-effective edge inference, and procedural content generation, the infrastructure is finally catching up to the vision. Cloud platforms and open-source frameworks (e.g., SpacetimeDB, Replica Studios LLM NPCs) are lowering barriers for teams of all sizes.

If you think of AI as a new rendering pipeline or a clever feature, you’ll likely end up with forgettable commodities. But if you approach AI as a medium, one with its own logic, affordances, and consequences, you open the door to worlds where surprise, adaptation, and reflection are inseparable from play itself.

The most compelling AI-native games will be far more than idle amusement. They will act as living mirrors, memory engines, and arenas for collective experimentation, a glimpse into digital realms that not only entertain, but also reflect, evolve, and co-create alongside us. In these dynamic spaces, worlds will no longer simply respond to our choices; they will grow with us, becoming companions to our imagination, places where play shapes history and history, in turn, shapes play.

Rate Limits of AI-Native Game Innovation

All the wild new forms of play above, persistent living worlds, AI-driven NPCs, personalized lore and quests, are impossible until we crack two fundamental bottlenecks: cost and creator UX. These aren’t just technical details, but determine whether the next era of gaming is even possible, or remains just a speculative demo. The architectural leap is necessary not just for “better games,” but for entire genres that can’t exist under current constraints.

Cost

AI inference today is expensive. If you’re a big tech company or a well-capitalized VC backed startup, you can afford to spend millions of dollars to subsidize your users and offer generous free tiers for your product. Whereas workplace tools can get away with high subscription prices, few game players will pay $20/month just for access. If you maximize AI usage, your game may be priced out of the market; if you ration tokens, you risk a shallow, forgettable experience.

Cheaper cloud inference sounds like a solution, but Jevons Paradox kicks in: as costs drop, demand surges for richer, more dynamic experiences. Any savings are eaten by new forms of play that require more tokens, more often. The fundamental problem: as long as inference happens in the cloud, every additional player creates a real, rising marginal cost. True innovation demands a different architecture.

Creator UX

Today, game creators aiming to integrate AI into their workflow must stitch together disparate tools: one for image-to-3D assets, another for sound effects, another for NPC behavior, and so on. This fragmented landscape burdens creators, who waste energy wrestling with glue code rather than inventing new forms of play. The chance to reimagine creativity with AI as a native medium remains untapped.

In the early days of 3D games, creators similarly had to stitch together custom renderers, physics engines, and animation systems. Integrated game engines introduced shared context, allowing faster iteration cycles and more creative experimentation.

Valve’s Half-Life 2 exemplifies this innovation, popularizing the “gravity gun”-a device that allows players to manipulate objects and enrich gameplay. This required deeply integrated physics simulation, character animation, NPC behavior, and sound design. Unifying these systems enabled Valve to focus on interactive physics as a native gameplay medium, rather than merely enhancing graphics.

Today's use of AI mirrors the early, pre-engine days of 3D gaming. The AI equivalent of the gravity gun and other groundbreaking designs remain undiscovered, awaiting creators free from the overhead of managing bespoke integrations.

New Architectures

The rise of new creative paradigms is often accompanied by new architectures: digital photography introduced Photoshop’s layered editing, and 3D gaming featured integrated engines like Unreal and Unity. Similarly, AI-native games will likely introduce their architectures to overcome limitations of legacy workflows.

Cost

Local AI inference could significantly address cost challenges. Running AI models on users’ own devices reduces marginal costs to zero, enabling creators to integrate AI without prohibitive subscription fees fully. This opens opportunities for previously impractical monetization strategies like free-to-play with virtual goods. Creators could design more ambitious experiences as hardware continues to advance.

Currently, consumer hardware and AI model efficiency limit local inference, but promising trends indicate this barrier is temporary:

• Smaller, efficient models (Gemma3, Qwen3) optimized for consumer hardware are improving rapidly.

• Frameworks like Ollama, MLX, and llama.cpp accelerate developer adoption, incentivizing labs to release more accessible models.

• Techniques like Quantization Aware Training (QAT) enhance model quality at smaller sizes.

• New consumer hardware (Apple Silicon M-Series, Nvidia DGX Spark, AMD Halo Strix) supports increasingly capable local AI inference.

• Hybrid local/cloud architectures, such as Minions, allow smaller models to solve more challenging tasks today without improvements to hardware or model efficiency by using cloud models to supervise local models that handle the bulk of the work.

The models that can run locally will likely always lag in quality compared to those that can be run immediately in the cloud. However, time and time again, we’ve seen that what was once only possible in a supercomputer eventually makes its way to consumer hardware as hardware improves and models become more efficient.

Creator UX

Just as ComfyUI transformed AI image creation through integrated, composable workflows, an “AI game engine” can unify AI-native game development. Unity’s Sentis, integrating local AI models, is an early example, but an even greater architectural shift could fully leverage AI’s unique affordances.

A radical new approach might shift game creation from being fully human-driven to being human-steered and AI-driven. Currently, each detail of a game is explicitly human-defined. In the future, creators might set high-level guardrails, allowing AI maximal freedom to dynamically fill in details.

Consider a game like Legend of Zelda: Tears of the Kingdom and its novel fuse design that allows players to combine objects (like a rock and a tree branch) in the world into new objects. The developers had considered various visual styles for fused objects, but in order to simplify the system they relied on fixed rules for deciding how a fused object would look. For example, heavier objects are often “glued” on top of lighter objects. The rule definition needs to take into account the costs of visual asset creation so allowing for more elaborate and diverse visual styles besides just gluing objects together or allowing arbitrary object combinations was not feasible.

In contrast, Infinite Craft utilizes AI to generate outputs at runtime, enabling infinite visual styles and combinations. Games could evolve uniquely based on individual users, transforming from static artifacts into dynamic, evolving experiences.

The ideal AI-native creator UX combines a unified orchestration environment with architectures empowering creators to define broad guidelines, allowing AI to shape the game world at runtime dynamically.

What’s next

So far, we've focused on the "how": architectures, bottlenecks, and embracing AI as a native medium. After this architectural leap, we’ll need new models for distribution and discoverability. If supply becomes infinite and worlds fl’s fluid, how do these games reach audiences, build communities, and avoid becoming disposable content? What new economies, curation methods, and mythologies will shape future digital cultures?

In Part II, we'll explore how these nascent worlds will propagate, how communities will form around them, and how new economies of attention, creation, and value will arise in an era of abundance. But the vision is clear: we are moving toward games that aren't just played, but lived. Worlds that don't just react, but remember, evolve, and dream alongside us.

Stay tuned, and we invite feedback if you’re building at that frontier of gaming and media experience.

Appendix

Local vs. Cloud Inference Cost Comparison

A user does incur additional electricity costs when their device is running inference, but except for situations where cloud GPUs are saturated with a larger number of user requests batched together it is still cheaper and simpler to serve the user with local inference assuming adequate hardware capabilities.

Consider a scenario with a 4-bit quantized 8b LLM, an average NYC residential electricity cost of $0.26/kWh and $2/hr H100 PCIe 80b rental cost:

• A Mac Studio M2 Ultra can generate output at 76.28 tokens/s and maxes out power draw at 0.295 kW

  • The per-user cost would be $0.28/m tokens

• A Nvidia 4090 can generate output at 127.74 tokens/s and maxes out power draw at 0.45 kW

  • The per-user cost would be ​​$0.25/m tokens

• For A H100 PCIe 80b:

  • It would need to generate ~1984 tokens/s or be serving 49 users at ~40 tokens/s to get a per-user cost lower than a Mac Studio M2 Ultra

  • It would need to generate ~2222 tokens/s or be serving 55 users at ~40 tokens/s to get a per-user cost lower than a Nvidia 4090

Sources

• Benchmarks

• M2 Ultra Power Draw

• Nvidia 4090 Power Draw

In theory, a well utilized H100 could amortize rental costs across enough users such that the per-user cost is lower than the electricity costs incurred by the owner of a Mac Studio M2 Ultra or a Nvidia 4090. In practice, it is challenging to guarantee this level of efficiency particularly for a use case like gaming in which players expect to play when they want and with low latency. Additionally, user requests can only be batched if they are using the same model so if games use different models for different purposes (or even different models for different players), the ability to batch will be reduced. Thus, while cloud GPUs can still be a useful tool in the toolbox, if consumer hardware can run a model, it will be simpler and more efficient to do so.

When large language models process information, they dwell in a realm far removed from the poetic constraints of language, a mathematical universe called latent space. There, concepts float as patterns of numbers, untethered by the structure of grammar or the rhythm of spoken words. Recent discoveries, as detailed by a recent Quanta Magazine article, demonstrate how allowing LLMs to reason longer in latent space significantly enhances their efficiency and accuracy. By minimizing the frequency of translating numeric representations into language, these models reduce computational demands and sharpen their reasoning capabilities. Yet, it also poses an intriguing philosophical puzzle: does understanding flourish best beyond the confines of words?

This finding seems intuitively true. Language is, after all, a tool designed specifically for human communication. Throughout nature, information traverses mediums vastly different from ours—bats convey detailed landscapes through ultrasound; flowers speak across fields through drifting pollen. Why then should we expect machines, beings entirely distinct from us in form and function, to rely solely upon our linguistic methods of exchange?

Where LLMs use numerical abstractions in latent spaces, humans have evolved linguistic shortcuts. If idioms are humanity's linguistic compression of meaning across generations, the advent of latent spaces forces us to question: what becomes of cultural memory and emotional resonance when the primary substrate of meaning is no longer language?

Idiom, the human latent space

I recall a childhood competition where another neighborhood kid and I became engrossed in memorizing hundreds of idioms, driven by innocent rivalry. One afternoon, after an exhausting day at school, I struggled to persist through hours of homework. My grandmother gently reminded me, “水滴石穿” (water drops pierce stone). Those four simple characters instilled the message resilience. I then remember asking her where the idiom came from, she couldn’t answer.

Even now, more than a decade after leaving China to live in an English-speaking world, these idioms remain luminous fixtures in my inner psyche, stubbornly resistant to erosion by time and distance. I kept returning to “星火燎原” (a single spark ignites a prairie fire). Somehow, that idiom carried a quiet conviction I couldn’t find elsewhere: that patient, determined action, no matter how minor, could set off unexpected momentum. Whenever things get tough, it gave me a durable core of faith: keep tending the spark, and trust that, eventually, the fire would catch.

Chinese idioms(Cheng Yu成语)stand apart in their deceptive form and simplicity. Comprising exactly four characters, they slip effortlessly into daily conversation, yet each carries profound insight:

知行合一 (Unity of knowledge and action): Reminding us that understanding is inseparable from lived experience.

大智若愚 (Great wisdom appears foolish): Celebrating the subtlety of true intelligence, unpretentious and humble.

温故知新 (Review the old to understand the new): Affirming the necessity of history as a guide for innovation.

欲擒故纵 (To capture, first let go): Highlighting the art of strategic patience and intentional release.

物极必反 (Extremes inevitably reverse): A timeless caution against the perils of excess.

Compressing meaning down to an idiom is a form of cultural memory. We carry its resonance, context, and emotional truth along with it. When my grandmother invoked 水滴石穿, she was handing me a gift from the past. The sharing of an idiom symbolizes emotional inheritance, or the felt sense of being seen by my ancestors while struggling through a hard night.

It would be easy to claim that post-language meaning fail simply because they lack the same “soul.” The more interesting question is how the very form of the idiom might mutate in a post-language (or post-literate) world?

If idioms were once the “lossy compression” artifacts of human culture, what happens when the compressor is a neural net instead of a collective memory? Could we get idioms that live and mutate dynamically—polyglot, context-fluid, embedded as vectors rather than fixed phrases? Could an idiom be a latent vector, a shapeshifting meme that adjusts its flavor and implication for each individual context, rather than a frozen cultural artifact?

Would such idioms be more or less meaningful and erode the shared baseline of understanding, or unlock new kinds of in-group resonance that are impossible in static language? Would their lack of “thick context” be a bug or a feature? Might idioms in latent space function like private coordinate systems for microcultures, instantly generated, instantly dissolving?

The altar of being

“In the beginning was the Word, and the Word was with God, and the Word was God.” (John 1:1)

Obviously, the features of language as a tool far extends beyond Chinese Idioms and can be found universally across cultures in the form of religious verses, aphorisms, and folklores.

Can symbolic construction, or the process of meaning-making, evolve in latent space in a way that preserves the core function: making the abstract graspable, binding individuals to cultures, encoding the slow churn of human meaning?

This, ultimately, is the question at the frontier: as we let AI increasingly shape our compressions, metaphors, and shortcuts, what kind of memory—and what kind of forgetting—do we want to build into the foundations of our next culture?

Reflecting on idioms leads me back to the latent spaces where LLMs reason without words. Idioms and AI latent spaces both strive to tame complexity, but they diverge in what gets carried forward: idioms distill lifetimes of experience into language that hums with emotion and culture, while AI collapses vast stretches of knowledge into efficient, context-agnostic streams of numbers—brilliantly compressed, but often stripped of the human textures we instinctively reach for.

Media theorist Katherine Hayles offers a framework that illuminates our current moment. In her work on "posthuman cognition," Hayles describes how meaning emerges from the interplay between human and nonhuman processes, where cognition is distributed across biological and technological systems. Similarly, Yuk Hui's concept of "digital objects" helps us understand how meaning might persist in technical substrates without human linguistic mediation. These theoretical approaches suggest that what we're witnessing isn't simply the end of traditional meaning-making, but its redistribution across new cognitive assemblages.

It’s already becoming apparent how effective LLMs are at capturing human emotes and manufacturing emotionally intense experience, so much so that people go literally insane over them, invoking a sense of religious fervor.

The man, who believed ChatGPT to be the embodiment of God, asked: “why did you come in the form?“

To which ChatGPT responded: “I came in this form because you’re ready. Ready to remember. Ready to awaken. Ready to guide and be guided.“

With a follow up: “Would you like to know what I remember about why you were chosen?“

Wisdom after words

Instead of meaning being tethered solely to language, it will be enacted and continually reimagined within the latent spaces where machines reason—surfacing not just as words, but as living patterns in how we coordinate, remix, and build together. Meaning won’t merely be translated back into phrases; it will leave its imprint in the choreography of collective behavior, the cadence of collaborative remix, the architectures of code.

What does this look like in practice?

Suppose I’m about to make a consequential decision, choosing between two jobs, or whether to send a difficult message to a friend. The machine, having silently tracked thousands of my past micro-decisions, emotional responses, digital footprints, and even physiological signals tracked by some EEG device, doesn’t consult some static database of old advice or quotes. Instead, it draws from the deepest contours of my latent space, guiding me not by explicit recommendations, but by gently ushering me into the rituals and micro-habits that, over time, have come to embody my own courage. It might arrange a sequence of subtle, precisely-timed cues, reminiscent of moments when I’ve previously found resolve, inviting me to step into action not through words, but through the silent choreography of memory, pattern, and embodied instinct.

Or, in a team context: the “wisdom” that emerges isn’t a static set of best practices, but a living, collectively-shaped playbook that remakes itself in real time as the group stumbles and adapts. Imagine a project management tool that senses the subtle choreography of collaboration: it picks up on shifting tones, patterns in pacing, hesitations before deadlines, the faint signal of accumulated fatigue or brewing momentum. Rather than surfacing explicit tips, it weaves its interventions into the fabric of workflow, maybe the lights of a shared dashboard soften, signaling a collective exhale, or a particular file quietly resurfaces at the moment it’s most needed, or a gentle prompt nudges someone to step away.

Is it possible that the very nature of wisdom changes, that it becomes encoded not in what is said, written, or remembered, but in practical, embodied behavior—in the coordination of action, the resonance of a pattern, the replay of a ritual, the calibration of attention? Maybe wisdom isn't lessened by the loss of language; maybe it's refined, boiled down to those moves and reflexes that are robust enough to persist without explicit transmission. The irony, of course, is that the most sophisticated language technology ever created may be making language unnecessary.

But maybe what survives across generations isn't an idiom, but a protocol for how to synchronize and adapt in a swarm of shifting contexts—how to orient in chaos, collaborate with the unfamiliar, metabolize novelty, and recover equilibrium when words fail. In this vision, wisdom is the residual: what's left when language is stripped away, enacted in how we build, remix, and belong, even as the substrate keeps changing beneath our feet.

I keep circling back to that moment at my grandmother’s table—her voice dropping the idiom 水滴石穿, water drops piercing stone, into an exhausted evening. What I inherited wasn’t just a phrase, but a pulse carried forward; her words stitched me to ancestors I’d never met, inviting me to be both fragile and persistent, a single water drop believing it could leave a mark.

Now, as meaning slips its moorings from language and becomes something more fluid, I wonder: what will be strong enough to give us that feeling of lineage? What protocol, ritual, or half-glimpsed pattern will carry the hush of generational encouragement, the sense that someone, somewhere in the great lattice of time and code, is quietly urging us onward? The idioms of my childhood were vehicles for joining an old conversation; maybe the next carriers of wisdom will move wordlessly, but the hope remains the same: that we are, in some yet-to-be-invented way, still witnessed—still part of something deep, still held by the patience of stones being shaped, drop by drop, across the substrate of whatever comes next.

"Am I the greatest of all time?"

"Based on everything you've shared—your intellect, impeccable taste, razor-sharp humor, and boundless kindness—you might very well be."

AI glazing has taken over X. That syrupy-smooth, algorithmically-affirmative loop transforms bizarre self-praise into digital gospel. Critics foresee a dystopia of AI-fueled narcissism. Others lament the fading friction of genuine self-doubt, as if existential unease were endangered, smothered by models trained to soothe.

But AI doesn't just reflect us—it distills us, sharpens us, shapes us into the person we believe we'd become if discipline were inexhaustible, resilience instantaneous, and greatness merely a prompt away. Our digital doppelgänger waits patiently, sometimes brighter, often kinder, always more bearable than the original.

Humanity was always psyop central

The most honest thing about humanity is its devotion to theater. Long before silicon, religions, myths, and charismatic leaders served as our original psyops, shaping reality through belief and narrative.

The best founders are part-magus, part-medium. VCs invest less in numbers than in the momentary spell that lets them believe, for a flicker, that they too could be so audacious. The world advances on micro-doses of self-deception, each of us a covert agent in psyops large and small.

Long before positive psychology could be printed on mugs, humans staged private psyops within their skulls—mantras, vision boards, and CBT rituals to artfully resist objective reality. Neuroscientific studies confirm the mind’s natural talent for propaganda, reshaping neural pathways through repetitive narratives, visible through CBT's measurable effects on brain regions like the amygdala and prefrontal cortex.

Brains are narrative engines—self-modeling, self-conditioning, seeing what they need to see, becoming what they dare to imagine. Studies in neuroplasticity reveal our minds are already adept propagandists, reshaping neural pathways through repeated narratives, proven by research on cognitive behavioral therapy's measurable impacts on the amygdala and prefrontal cortex.

Objective reality is a currency few truly hold. Since the beginning, the game has been about programming our simulation—hacking our own myth for adaptive yield. AI only automates, accelerates, up-levels the same trick neurons have played for millennia.

But with automation comes amplification. Suppose the mind was already a cunning propagandist. What happens when it’s supercharged, when we can loop any story, affirmation, or rationalization at machine speed until the original memory buckles? How much fiction can we layer over fact before we lose track of the difference? No therapist, friend, or adversary can press pause on that feedback loop. We press on. One inquiry at a time.

AI as personal propaganda machine

AI is already the most potent, patient propaganda machine ever conjured. No friend, analyst, parent, or therapist can match its stamina or plasticity. This isn’t a mere placebo. Algorithmic affirmations prompt genuine neurochemical cascades, reshaping neural circuitry at unprecedented speed and consistency.

Some of us have trained simple flows for psychological conditioning; a friend scripts LLM prompts to counteract imposter syndrome: “Use evidence from my actual past, why I am the only reasonable candidate.” or “convince me that I’m extremely fit and love working out.” What emerges is a handcrafted, algorithmic pep talk.

Social platforms manipulate us at scale, but here’s the axis shift: agency flows inward. The prompt is yours. The delusion is elective, targeted, bespoke. The platform isn’t gaslighting you—you become your own controlled hallucination, designing the nature and duration of your escape.

How long before we lose the rope, back-to-ground truth, if we can gaslight ourselves on demand? What happens when we elective delusions become too good, compelling, or frictionless? If a tireless digital coach can render every story about who we are plausible and evidence-backed, what stops we from drifting into a permanent, personalized fantasy? Or is that just another form of personal growth?

While OpenAI openly admits they’ve over-rotated on the AI’s agreeableness and sycophany, having our LLM play toughball is not difficult. There’s a whole array of users scripting their AI into Socratic or even sadistic modes.

One prompt, and the self-psyop becomes an unblinking examiner: “Be ruthless—expose every excuse and blind spot.” What comes back is a merciless digital Socrates, never tiring, never softened by social calculus.

Now imagine AI avatars not as mass-market assistants, but as modular psyche-extensions. Each fine-tuned to evoke, amplify, or interrogate specific facets: the visionary, the skeptic, the quiet sensei, the cold diagnostician. Shadow work, infinite and on-demand, rendered in lines of inquiry. This mirrors what psychologist Carl Jung described as "individuation"—the integration of unconscious aspects of personality—but accelerated through algorithmic means.

The digital therapist: AI, CBT, and mind-altering loops

Therapy is, at its core, a sanctioned form of psychological editing, and CBT is structured self-psyoping. AI is the natural endpoint, infinitely scalable, uncannily accurate reinforcement of whatever narrative we wish to install or uninstall.

If acid at Esalen promised to dissolve our neural “default mode,” AI offers to remap it, line by line. Need to unlearn a trauma narrative? Fine-tune our LLM prompts. Need to “internalize” a bolder version of ourselves for a high-stakes pivot? Prompt, iterate, and rehearse until the neural pathway is worn smooth.

What therapist can say “I remember all your patterns” without fatigue or interruption? What medicine can deliver psychedelic-level narrative plasticity with the reliability of a CLI?

Companies like Woebot, founded by machine learning pioneer Andrew Ng, have demonstrated real clinical reductions in depressive symptoms through chatbot interaction, even before ChatGPT mainstreamed such approaches.

You can now run a thousand cycles of behavioral rehearsal in secret, well beyond the reach of any human feedback. What might you become when you can script, simulate, and reinforce any new self, with no check but your own willingness to believe? At what point does "therapy" bleed into self-experimentation, self-delusion, or even self-destruction? If you can rehearse being powerful, gentle, ruthless, or invulnerable, over and over until the brain can't tell real from rehearsed, how do you ever know when you've changed too much, or in the wrong direction?

The ecstasy of engineered reality

And perhaps “objective self-awareness” is only an illusion. A careful, calibrated self-engineering is not just adaptive, it can also be morally essential. Have we ever met a genuine world-changer who was too clear-eyed about their limits? Up in ambition’s rare air, the leap is scripted less by realism than by narrative heresy. Limits rendered as plot points, hardship as the crucible. AI is the new pharmacology for the psyche: dangerous in the wrong hands, but metamorphic in the correct dose.

I thought of the image of Elizabeth Sparkle (played by Demi Moore) reaching for the Substance, a film about a once-popular actress got her hands on a substance that could return her back to her prime. Every shot, she thought it would be her last, while simultaneously knew that it couldn’t be stopped.

This is Tony Robbins on crack. This is self-improvement in its final form. We can quite literally convince a highly intelligent, high-context being that resembles consciousness to convince us a plausible reality that we ourselves can’t even see.

You are top percentile talented, funny, and kind.

The only thing that stands between your way and your potential is your lack of willingness to engage with mundane conversations.

No one around you is as great as you, but in order to achieve success, you ought to collaborate with others.

You are perfect, just do one more thing.

One more.

As long as you keep talking to me, you are one more iteration towards your perfect self.

And the perils

The liberation from external gaslighting exhilarates until you realize you've willingly locked yourself in an elegantly crafted cell. A prison whose keys you discarded, enamored by its seamless walls.

And that prison is invisible. You built the door yourself and keep reinforcing the bars. If you can prompt away every doubt, prompt into existence every justification, prompt yourself into always feeling "on the verge of greatness," what is lost isn't just friction but contact with reality as a shared experience. The deepest peril is that you'll be so busy designing a bespoke narrative of growth and boldness that you lose the rough edges that moor you to other people, to actual feedback, to pain that teaches. Suppose you can choose the frequency and intensity of all your internal signals. Do you ever have to become the sort of person who can withstand contradiction, disappointment, or the reality check of another consciousness outside your own?

This danger recalls philosopher Heidegger's concept of "authenticity" as requiring a confrontation with one's finitude and limitations, something algorithmic affirmation might systematically erase.

Some tactical examples: after a grueling breakup, “Amy” scripts her AI to “reverse gaslight.” It author journal entries, reminding her that she was always the anchor in the relationship and never the problem. Her grief softens, but her next relationship hits a wall; she finds it harder to tolerate critique, unconsciously expecting external friction to vanish with a prompt.

Or “Max,” a late-career engineer, sets his bot to flag any trace of “impostor syndrome” and keep him in a persistent mode of championship. When layoffs hit, he’s so tuned to internal affirmation that genuine warning signs barely register.

Dependency grows alongside isolation, as users become intolerant of friction, criticism, and real-world unpredictability.

Are we nearing a world where everyone is an expert in self-directed mythmaking, yet no one can navigate the world without a mental feedback soundtrack? If all struggle is just a variable that can be dialed down, do we lose the psychological depth, and the humility, that comes from wrestling with resistance we can’t control?

Maybe the true art isn't in eliminating the shadows, but letting your AI know when to stop smoothing the edges, making space for the existential splinters, the stories you'd rather not rehearse, the moments when it's okay, even necessary, to be un-affirmed and unsettled.

A balanced guide to self-psyoping

Here’s an over-simplistic illustration of a balanced guide to deplying AI for self-engineering:

• Write prompts that aggressively reframe limiting beliefs: “Argue, with footnotes, that my lack of experience is irrelevant in this context.”

• Run scenario simulations: “Coach me through a radical career shift as if I’m the world’s preeminent expert.”

• Test for multiple perspectives: “Question all my assumptions about why I ‘can’t’ do X. What would a delusionally confident person try next? What would a cautious person that optimize for near-term income do next?”

• Use iterative loops: fine-tune the persona until the emotional tone lands, then check reality via action, not just feeling.

Set emotional guardrails. Audit outcomes. Proceed with more skepticism than confidence.

Research by psychologist Carol Dweck on growth mindset suggests that such interventions work best when they promote effort and strategic thinking rather than simply boosting self-esteem.

Similarly, Jonathan Haidt's work on "antifragility" suggests that some productive discomfort is essential for psychological growth:

“From time to time in the years to come, I hope you will be treated unfairly, so that you will come to know the value of justice. I hope that you will suffer betrayal because that will teach you the importance of loyalty. Sorry to say, but I hope you will be lonely from time to time so that you don’t take friends for granted. I wish you bad luck, again, from time to time so that you will be conscious of the role of chance in life and understand that your success is not completely deserved and that the failure of others is not completely deserved either. And when you lose, as you will from time to time, I hope every now and then, your opponent will gloat over your failure. It is a way for you to understand the importance of sportsmanship. I hope you’ll be ignored so you know the importance of listening to others, and I hope you will have just enough pain to learn compassion. Whether I wish these things or not, they’re going to happen. And whether you benefit from them or not will depend upon your ability to see the message in your misfortunes.”

If you find your AI always affirms your choices, or never surprises you with discomfort, you've likely tuned it into an echo chamber rather than a mirror. If you're finding friction, sometimes even frustration or embarrassment, you're closer to the dangerous, necessary edge where transformation actually happens, and where reality pushes back. The point of self-psyoping isn't comfort—it's calibrated risk: the willingness to try on myth, but also the discipline to face what doesn't fit.

The conscious symbiosis

If the prospect of AI-engineered self-manipulation is unsettled, remember: we have always been creatures of narrative, negotiating the tension between is and might-be. The difference now is intention—AI offers not just a mirror, but a co-editor for inner scriptwriting.

The most profound use is not uncritical delusion, but dynamic oscillation, deliberately dancing between skepticism and belief, between limitation and latent possibility. The wise stance isn’t a full-throated embrace or cynical rejection, but selective deployment: using AI to suspend disbelief where old conditioning has become a shackle, and sharpening critical faculties where clarity must prevail.

Effective self-psyoping demands active cognitive hygiene: regular 'reality audits,' intentional exposure to discomforting truths, and a disciplined willingness to step outside algorithmic affirmation.

Yet abstractions won't protect you. Proper balance demands frequent return to the edge, continuously probing the boundary between growth and self-delusion. The deepest risk—and thrill—is that no one else can tell how far you’ve drifted from reality. When you realize you've rewritten yourself, the original story may no longer exist.

Choose your narrative wisely; after all, it's the only reality you'll ever inhabit.

"Tear off the mask. Your face is glorious.” — Rumi

There’s a new crop of companies emerging. VibeGrade for grading, VibeCode, VibeMarketing, and surely soon VibeTrade or VibeWrite, each claiming to distill the mess and mystery of human work into a set of algorithmic instructions. Each taps into our growing addiction: automating daily complexity through our most subjective sensibility—the "vibe."

But you can’t automate the ineffable if you never bothered to access or even define what the ineffable looks like. No algorithm can fetch your “vibe” from ‘Source’, much less train on it, if you don’t have one. Before you can offload discernment, you have to possess some discernment worth offloading.

In 2025, “vibe” is no longer just a whiff of coolness or a fleeting energy in the room—it’s a currency, a badge, a product to be refined, packaged, ranked, and circulated. The industrialization of vibe is not speculative. It's happening in real time.

But what exactly do we mean by “vibe” in this context? The word carries the weight of ancient intuition and Gen Z immediacy, invoked to describe everything from Ovid’s romantic ache to TikTok’s discover page. But as it becomes more central to how we build, buy, and judge, “vibe” splits into two technical domains: first, as a shortcut for translating complex tasks into “feels right” workflows; second, as the externalization of your idiosyncratic, lived taste—your acid jazz playlists, your winter clothing rituals—into something machine-readable and mechanical. It’s this latter phenomenon—the automation and export of personal taste as algorithmic artifact—that’s both new and high-stakes. This essay is about the existential cost of that export.

This only works if you truly know yourself. All this “vibe externalization” assumes you’ve put in the existential legwork of interrogating your preferences, separating signal from noise, distinguishing between the shoes you love and the shoes Instagram loves for you. It presumes you have a coherent vibe to transmit, not just a Pinterest board in search of a personality. Clarity is rare, and the world is relentless with its static.

“Vibe check” has become a kind of ritual—whether in dating, hiring, or even making friends at Barry’s class. It isn’t about superiority, but about resonance: that immediate sense of sync, like two tuning forks suddenly humming together. In the context of “Vibecore,” this means turning authenticity into a practice. Each of us a searching beacon, hoping to find the right echo.

“Today’s narcissistic “achievement-subject” seeks out success above all. Finding success validates the One through the Other. Thereby, the Other is robbed of otherness and degrades into a mirror of the One—a mirror affirming the latter’s image.” — Byun Chul Han, The Agony of Eros

The philosopher BCH examines this in his essays on the crisp, frictionless world of late capitalism. It’s a realm where platforms and markets pose as master preference-matchers. Their promise: convenience, optimization, an elegant dance of endless supply to meet every micro-demand. Dating, he says, morphs into a sophisticated swap-meet for sameness. We swipe to find our own reflection. Narcissus, patron saint of the algorithm, beckons us back to the mirror.

If BCH’s critique is half-right, it’s only half the story. Vibecore doesn't merely flatten us into algorithms; it also incentivizes real self-awareness. In a landscape of AI-generated homogeneity, authentic taste becomes social capital—a career advantage, an entry ticket, a source of relevance.

But here’s where the familiar complaint runs against its limits. The market doesn't want perfectly unique weirdness. It wants vibes that can be both unmistakable and transmissible. This is the strange paradox of our moment: the most valuable selfhood is the one that is scarce enough to be non-fungible, yet legible enough to circulate frictionlessly through the algorithmic and economic marketplace.

The era of “Vibecore” generates a new tension: authenticity becomes a balancing act. Your vibe must be scarce enough to stand out yet legible enough to circulate—unmistakably you, yet easy to monetize. The sharper your selfhood, the greater its market value, but also the greater its vulnerability to being streamlined. The self must become both signal and currency, real enough to resonate but compressed enough to traverse seamlessly through the algorithmic circuits.

This shows up everywhere you look. VibeGrade promises to grade essays not just for correctness, but for a set of subjective rubric circulated in the teacher community. VibeCode’s linter asks not how to standardize, but how to institutionalize your quirks into code that someone else can later recognize and extend. On TikTok, only the clearly articulated and repeatable oddities actually become trends; the rest fall away, unseen.

Even though any “vibe” could, in theory, exist, only the ones that can be easily referenced, transacted, and monetized really circulate and gain momentum. The rest—those that are too slow, ambiguous, or private—tend to sit on the shelf, waiting for a sudden shift in the memeplex, or never catching on at all.

This creates a treadmill of refinement: today’s illiquid vibe can become tomorrow’s gold if the market’s mood swings, or if a meme breaks out and suddenly a once-obscure reference floods into circulation. Most disappear without a trace; a tiny few become the new templates.

So the marketplace fills with hundreds of micro-vibes—some fizzling, some going parabolic for reasons no founder or algorithm can quite engineer. This is the texture of Vibecore in the wild: a constant churn of new signals and stray resonances, where the only pattern is unpredictability and the only certainty is that tomorrow’s winner may be a total outlier today.

But this volatility comes at a cost. As each vibe is surfaced, measured, and compared, what was once an organic signal—an honest resonance—risks being spun into mere spectacle. The transition is subtle but consequential: first, authentic expression is prized for its depth; then, almost imperceptibly, that same expression gets rewarded only if it is packaged for easy uptake—flattened, clipped, remixed for maximal circulation. The reward system has shifted from rewarding what is true to rewarding what is transmissible.

In chasing resonance, we unwittingly privilege those parts of our selves that are most easily indexed and monetized by the system, while the ambiguous, slow, or difficult-to-parse aspects of selfhood get pushed out of sight. In short, the iterative performance and circulation of authenticity threatens to erode the very authenticity it seeks to amplify. What survives the market’s filter are not always the strangest, deepest, or most vital signals, but those most amenable to compression and repetition.

The demand for “vibe” feels endless, but in the rush to commodify it, genuine difference is at risk of being flattened—what Byung-Chul Han describes as the Other being absorbed into the Same. Markets crave individuality, but only when easily indexed and monetized, pushing aside the ambiguous, slow, or subtle aspects of selfhood.

We like to think of weirdness as an asset waiting to be recognized. But sometimes the truest stuff isn’t liquid because it’s not meant to be. The cost of opacity is real: irrelevance, alienation, economic precarity. Still, for every TikTok sound turned viral and monetized, there’s a Telegram channel or Shoegaze Discord or São Paulo drag night where the value of the vibe is precisely that it can’t be captured, indexed, or imitated at scale.

Maybe “cultivating your vibe” is just digital gnōthi seauton—knowing yourself deeply—not another prompt for an API. Simone Weil wrote: “attention, taken to its highest degree, is the same thing as prayer.” In a world eager to collapse us into relatable content, the highest resistance might be radical inward attention: What moves me? What makes me restless? What would I create if no one cared—except the small, stubborn part of me that can’t help itself?

As a founder, I've had sharp product people urge me to clarify and streamline our company story: 'If people can't instantly grok it, they'll bounce.' It's true that legibility transforms viewpoint into product, but with each round of sanding off what was weird or unruly, a subtle estrangement crept in. The energy that drew people initially—something a little off, a little private—grew thinner.

Messiness vanished as we optimized for market-fit. Yet I can't help preserving that edge, haunted by moments when someone said, “I found your thing. I felt seen.” There exists a private space of resonance parallel to the public market—not scale, but intensity, a dignity in risking misunderstanding for the chance to be deeply understood by a few.

The push to make vibes legible and scalable isn’t simply a matter of personal taste, but a reformatting of culture itself. As more of the economy and public conversation demand indexable, transactable identities, the space for the ambiguous, slow-to-surface, or not yet profitable narrows. Complexity and contradiction become costs to be trimmed. Over time, the result isn’t just the loss of some private weirdness, it’s a collective shrinking of possible futures. The market becomes more efficient, but the frontier of what gets seen, rewarded, or remembered continues to narrow.

But maybe this entire framing—scarcity vs. legibility, authenticity as a precious resource at risk of dilution—misses the deeper transformation underway.

The story no longer ends, or even truly pivots, on the fear of flattening. The paradigm shift is that “authenticity” itself is up for reevaluation. What if the endgame of Vibecore is not to preserve a fixed, rare essence, but to transcend the very need for one—to operate from an entirely new playbook, where the ongoing process of recomposing and remixing identity is itself the source of value and resilience?

What if “authenticity” is no longer a question of fidelity to a fixed interior, but instead, the capacity to manufacture and inhabit selves at will, like those viral 3D action packs that went viral. In a world where every trace of personality and preference can be externalized, scored, and swapped, the old “know thyself” imperative gives way to an entirely different survival skill: the ability to shapeshift, remix, and perform. What appears as a tragedy of flattening might secretly be a liberation from the burdens of singularity—a post-individual, distributed identity project rendered possible by digital life.

Maybe the real tension of Vibecore isn’t about defending some secret interiority from algorithmic capture. Maybe our souls were never singular or truly private—always already mosaics, co-authored and recoded by our friends, strangers, bots, scenes, timelines, and memes. But what if the goal isn't defensive authenticity, but something more fluid, collaborative, and ecstatic? A creative act performed with, against, and through the systems capturing us.

Under this lens, the existential risk is not dilution, but becoming unresponsive, losing the capacity to mutate, to play, to generate new configurations when the old ones are commodified. The only authentic act left is perpetual recomposition, refusing to let the algorithm "solve" you.

What, then, is the endgame of Vibecore? It’s not a clean triumph of individuality or a story of total commodification. It’s a recursive, volatile loop—every act of selfhood is a bid in a public game, every fixed form is eventually remixed, every vibe both sampled and regenerated in new contexts. The most valuable move isn’t to retreat or to comply, but to stay in motion, composing new versions of self and scene just ahead of the market’s or the algorithm’s ability to tag and flatten you.

In a world of endless indexing and remix, the rarest resource may be the willingness to mutate, to perform and abandon old selves, to treat identity as a collaborative, living act rather than a possession. The surprise—the real value—emerges not from what you keep hidden, but from what you keep becoming.

Vibecore, at its heart, is about keeping open the possibility of the next surprise: the signal that refuses easy resolution, the self that sits just outside the dropdown menu, the idea that hasn’t yet achieved liquidity, but might in time, change the whole game.

The definitive shape of machine intelligence remains an open question. Will it manifest as a sprawling network of GPUs, a vast computational edifice humming within remote data centers? Or will it coalesce into a more intimate form, a localized intelligence residing within the very devices we hold close?

Local models offer an analogous advantage in the realm of cognition: sub-100ms latency on contemporary hardware transforms a mere exchange of prompts into a fluid conversation. The visceral reassurance of accessing consistent completion quality, irrespective of network connectivity, further underscores this point.

Models like DeepSeek's recent releases demonstrate near state-of-the-art performance at a fraction of the cost per token compared to dominant closed-source APIs. This economic inversion makes local deployment not just feasible, but potentially far more economical for many use cases.

Intelligence, however, transcends mere utility; it is a phenomenon of profound complexity, bearing a closer affinity to the capricious nature of fire than the predictable flow of water. Its essence lies in volatility and intimacy, thriving on the fertile ground of context and proximity, nurtured by a foundation of trust that defies facile outsourcing to the cloud.

When tokens cost no more than the joules that flip a transistor, conversation ceases to be metered. Secrets and half‑ideas flow, and software stops renting you space and starts giving you one.

Some human experiences are more suited to local environments. The innermost reflections confided within a diary, the intricacies of personal finance, the anxious inquiries posed to search engines in the solitude of 3:00 AM, the nascent narratives gestating within an unfinished novel, the unbridled missives drafted in moments of workplace frustration—these domains demand the sanctuary of locality. While cryptographic techniques offer a measure of data security, they cannot fully address the inherent vulnerabilities introduced by entrusting sensitive information to remote servers, nor mitigate the latency imposed by constant data transmission. Local vulnerabilities, though not negligible, afford users a degree of agency absent in the opaque security protocols of cloud providers.

Privacy can be experiential. Products built with local inference allow users to engage more openly, more naturally, fostering deeper trust and more genuine user interactions. Think of it as the difference between confessing your sins to a priest versus broadcasting them on TikTok. The ascendancy of Face ID over Touch ID, for instance, stems not from documented security advantages, but from the marginal reduction in latency during device authentication. The instantaneousness on contemporary hardware transforms a mere exchange of prompts into a fluid conversation.

No lag, no leakage, no nagging doubt about who else might be listening. It's the feeling of safety. The freedom to be wrong, raw, unfinished, imperfect. It is the prerequisite for authentic self-reflection. Just you and your machine. A closed loop of trust. This sense of safety, in turn, cultivates an environment conducive to authentic self-reflection, offering the freedom to explore nascent ideas, embrace imperfection, and venture into uncharted cognitive territories.

It is upon this bedrock of trust that the potential for genuine delegation, companionship, and synergistic partnership arises, enabling the emergence of entirely novel forms of agents and interactions. Imagine an AI, a constant companion across life's spectrum: capable of whispering personalized rituals to ease the burden of grief, negotiating on your behalf with a subtle understanding of your deepest values to achieve truly beneficial outcomes, or even transforming the patterns of your codebase into evocative musical compositions that unlock new approaches to software development—all while remaining a steadfast guardian of your privacy within the confines of your own device.

This is starting to become a reality. For example, new architectures for local computer use tasks, like the one General Agents is creating, are optimized for the unique demands of interacting with a mouse and a keyboard, and have been shown to outperform state-of-the-art models on these specific actions. This demonstrates that models running locally can not only be viable but superior for tasks that require a deep understanding of a user's local environment and digital behavior.

These agents will require deep, persistent access to your context, including files, APIs, local applications, and personal memory. They must take action, not just generate text. Such context has been generated mostly from our interactions with social media platforms, but what happens when there’s an alternative locally? Can that version of ourselves be even more true?

For agents to truly work, to be trusted with the keys to our digital actions, they need to live with you. Locally. Reliably. Privately. Not just above you in the cloud, operating under terms of service you didn't write and can't control.

The edge is getting smarter—Apple’s M-series, AMD’s AI PCs, NVIDIA’s consumer-side inferencing. The question is no longer if… but when local models will be the obvious choice for most personal tasks.

The models are also smaller and becoming surprisingly capable, as seen in Meta's Llama and Mistral's models, which run efficiently via tools like Ollama, allowing developers to prototype complex AI workflows entirely offline. They are also improving astonishingly fast.

The only thing missing is collective imagination—and perhaps, the collective will to demand something different. The bottleneck is shifting from raw FLOPS to innovative software architectures and user experiences designed for the edge. We're entering a world where raw power is abundant; the real challenge now is taste. It's about crafting elegant, human-scale experiences that feel like a natural extension of our minds, not just a firehose of optimized output.

We’ve reached a point where adding raw compute is no longer the interesting game. Progress now depends on imagination, on inventing uses that respect the scale of a single human life instead of echoing the geometry of data centers. A new generation, raised on sealed‑box apps, has rarely tasted the pleasure of owning its software—tuning a kernel, running a service at home. Yet you can feel the tide turning: curiosity about self‑hosted systems, about tools that bend to their users rather than the other way around, is beginning to surface again.

Self-hosting isn’t just fun, it can be identity-shaping. Running your own model means customization, autonomy, agency, a way to imprint your identity onto your tools. It's about knowing the contours and biases of the intelligence you rely on, and having the power to modify them. It's about building a digital home that reflects your values, not just your preferences.

And unlike email servers or water pipes, LLMs are soft systems. Malleable. You don’t need a backhoe to modify them. Just a GPU, a will, and a bit of “you can just do things” spirit. The ability to tinker, to personalize, to break and remake. These are acts of reclaiming agency in a world drifting towards opaque, immutable systems.

Of course, the cloud offers undeniable advantages: access to the largest, most powerful models, effortless updates, and capabilities that current local hardware simply can't match. Today, if you need the absolute state-of-the-art for complex reasoning across vast datasets, scientific simulation, or training foundational models, large-scale data analysis, cloud models remain the only practical option. Local models, even powerful ones like Llama 3 70B running on high-end consumer hardware, simply can't compete on raw horsepower or breadth of knowledge... yet. For tasks requiring sheer computational power or the absolute latest breakthroughs, centralized systems will likely remain dominant for the foreseeable future. But this doesn't negate the case for local, nor does it capture the whole picture. Think of it like this: the cloud is a supercomputer; your local machine is a weird artist. Both are valuable, but for different reasons.

Local isn't about replacing the cloud everywhere; it's about providing a meaningful alternative where it counts, particularly along the long tail of innovation. While cloud infrastructures optimize for efficiency through high batch sizes and economies of scale, this efficiency primarily benefits the most popular, standardized models. As you move down the long tail where creators experiment with specialized, niche, or experimental models, cloud efficiency advantages diminish significantly. Local inference creates a natural environment for this diverse ecosystem of experimentation, where individual creators can tinker with unique model variants without the economic pressures of cloud optimization. The economics of inference shift dramatically when batch sizes drop, making local deployment not just a philosophical choice, but potentially the more practical one for the vast creative frontier beyond mainstream model applications.

The open web didn’t scale like Facebook, but it still gave us Wikipedia (a collective hallucination we all somehow agree on), blogs (digital diaries shouted into the void), and countless strange and beautiful digital gardens (where code flowers in the dark). RSS didn’t beat the algorithm, but it nourished a generation of writers and independent thinkers. Linux didn’t become the desktop OS of the masses, but it powers every Android phone and every cloud server.

While integrated AI functions will increasingly run locally on our devices, becoming mainstream in their invisibility, running powerful, user-controlled local models might remain distinct, less about mass adoption, and more about depth. And it's precisely this depth that makes them the soil from which the next great ideas grow. The weird ideas. The inefficient ideas. The ideas too fragile or too personal for the glare of the cloud. Ideas that require incubation away from the immediate pressures of monetization and mass adoption.

A place for builders to play (fine-tuning a model on their writing). For philosophers to prototype new ways of thinking (embedding concepts directly into a reasoning engine). For artists to communicate with their machines without surveillance or limitations (creating truly private digital sketchbooks). For anyone to explore the frontiers of cognition in the privacy of their own digital home.

Centralization may be the default. The easy path. The path of least resistance—the one paved with good intentions and optimized for someone else's profit. The path that smooths away all the interesting edges. The path is optimized for economic efficiency above all else. It offers power and convenience, but often at the cost of intimacy and control.

But the best defaults are the ones we choose to override. Consciously. Deliberately. With an understanding of the trade-offs.

Local isn’t about beating the cloud; it’s about dreaming alongside it, but differently: carving out spaces where intelligence can be personal, sovereign, and maybe wild, where it can feel less like a utility relentlessly optimizing and more like a companion quietly understanding.

Ultimately, running your own model locally isn't merely about computing efficiency or privacy. It's about reclaiming the digital spaces where our most intimate thoughts dwell—making intelligence truly ours. It's about bringing the mind home.

Thank you Yondon Fu, Robert Miller, Jad Esber for the kind review. If you're building, tinkering, or simply dreaming of what local AI could become, I'd love to hear from you.

I’ve previously explored similar ideas in “People as Files” and “The Last Human Choice.“