The Art of Understanding What's Going On
The survivors will be those who've meticulously deciphered the underlying mechanics, exploiting subtler truths while others chased fleeting illusions.
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.