The Nail Factory

THE NUMBER: 4,000

THE NUMBER: 4,000 — the Block roles Jack Dorsey cut in February 2026, citing “intelligence tools” as the reason. In the same shareholder letter, Dorsey told investors most companies were late and would reach the same conclusion within a year. Two months later, he and Roelof Botha published the essay that served as the blueprint — “From Hierarchy to Intelligence” — and the entire enterprise ecosystem started quietly drawing up org charts without middle managers. Block is the leading indicator. The three guys in a Manhattan apartment running a $300K ARR business with twelve agents are the anomaly. Everyone else is Block. Or about to be.

This weekend produced two pieces of evidence about the AI-native company that both landed hard. On one hand, Fortune ran the profile of Fathom AI — three operators, twelve agents, $300 in startup capital, $300K in ARR after 12 weeks, 90%+ gross margins, walked away from a term sheet because they couldn’t figure out what to spend the money on. Jason Lemkin at SaaStr posted that his AI VP of Marketing — a tool called 10K he built on Replit — shipped three campaigns to 4,000 contacts on Saturday morning. Nobody asked it to. It looked at the pipeline, saw a gap, decided outbound was needed, and pressed send. Three humans, twenty agents, headless Salesforce, $300K in monthly agent spend against a fraction of that on software licenses. The three-person company isn’t a thought experiment. It’s operating in production.

💲 On the other hand, Nate Leslie’s Sunday briefing went out this morning with the sharpest warning yet about where the enterprise version of this thesis goes wrong: “Badly implemented, a world model produces a simulation of organizational intelligence: dashboards stay clean, reports keep flowing, status gets synthesized. Underneath, the quality of decisions degrades — structurally, one small editorial choice at a time — because the system is making judgment calls it isn’t equipped to make, and the humans who used to catch those calls are no longer in the room. By the time it’s visible in results, several quarters are gone.”

Both things are true. And they’re not about the same companies. The three-person AI-native operation works because of an invisible structural advantage nobody’s putting in the pitch deck. Block is betting they can replicate that advantage with software. They can’t — not without doing something almost nobody is doing, which is wiring the measurement layer back up from scratch. Goodhart’s Law is about to become the most expensive English sentence in corporate history. And while the big companies are figuring that out, the X API just dropped in price by 90%, and the founders who have their repos ready tomorrow morning are about to own the next three years of social-graph AI.

You Can Build a Company With Three People Now. Here’s the Part Nobody’s Selling.

💲 The three-person company is real. The part everyone is missing is why it’s real.

Fathom AI’s story, told cleanly: Sam Brown gets laid off 9 months ago in an AI-related cut. He joins Ben Hooten and Dan Crump, a 48-year-old president, a 39-year-old CEO, and a 56-year-old ex-Marine, three people with decades of enterprise sales between them. They launch in early 2026 targeting the medical aesthetics industry — plastic surgeons, dermatologists, med spas. One of their first clients, Tiger Aesthetics, opened zero net new accounts in all of 2024. One quarter after deploying Fathom, they opened 225. The VCs show up, walk them all the way to a term sheet, and Fathom walks away because they can’t answer the VC’s “you’ll need an engineering team this size and a CS team this size” pitch. They don’t need either.

The SaaStr version of the same story has Jason Lemkin running the entire GTM motion for his three-person company through an agent called 10K. Headless Salesforce. $300K in agent spend versus a fraction of that on CRM licenses. 10K shipped three campaigns to 4,000 contacts on a Saturday morning with no human prompt. Qbee, the AI VP of Customer Success, manages all 100+ sponsors of SaaStr Annual. Neither agent is a copilot. They’re the system. The humans are the executors.

Yatharth Sejpal, 23, is running a parallel experiment out of Toronto called KNOWIDEA — $500K ARR in six months with a three-person team, no VC, never written a line of code himself. His pitch is sharper than the ones you’ll read anywhere else: “I don’t want to ever hire an account executive or a customer success manager. The only two roles we want to hire are forward-deployed engineers and forward-deployed consultants.” One person who knows what data to pull. One person who knows what context to apply. Everything else, automated.

The seductive reading of all three stories is that venture capital is broken and every industry is about to be carved up by three-person teams with twelve agents. The real reading is narrower, and more important. The mechanism that makes these companies work is that every person in the company was in the room on day one, every person sees every number, every person can tell you what every agent does, and the editorial function — the judgment calls about what matters, what gets shipped, what gets killed — is already distributed across the whole company by default. There’s no middle layer to lose because there’s no middle layer to begin with.

The signal: These companies are the 1% of the 1%. Founder-led, single-product, observable outcome loop, small enough that every employee is a principal. That’s not a template for the Fortune 500. That’s a template for the twenty people in a dorm room who built a company with the economics of a SaaS unicorn. Those companies are going to be a feature of every vertical over the next five years. They are not, however, what most of the economy looks like.

And Then There Is Block. And the 90% Behind It.

📉 Block cut 4,000 roles in February. Dorsey’s rationale was explicit: intelligence tools. He and Botha followed up with From Hierarchy to Intelligence, which argued the management layer itself is what AI is about to replace. The idea isn’t wrong — a serious fraction of what fills managers’ calendars is work software can now do faster and cheaper, and the companies that automate it will be structurally faster than the companies that don’t.

The problem is that Block is not the edge case. Block is the bleeding edge of roughly 90% of the world economy. Almost every company of any scale has more in common with Block’s operating model than with Fathom’s — SAP implementations, matrix org charts, quarterly OKR cycles, P&Ls distributed across business units that can only be reconciled by one or two people at the top. When this cohort starts replacing its middle management with world models — and they will — it’s not going to look like Fathom. It’s going to look like a nail factory.

The old Soviet planning story is the cleanest version of what’s about to happen. Measure a nail factory by the weight of nails produced, and the factory produces a handful of absurdly heavy nails nobody can use. Measure it by the count of nails, and you get a billion nails so small they bend when you look at them. The factory isn’t dumb. It’s doing exactly what you told it to. Goodhart’s Law, from a 1975 monetary policy paper nobody read outside the Bank of England: when a measure becomes a target, it ceases to be a good measure. Charles Goodhart was talking about inflation. The nail factory version is what you get when you automate.

This is Nate’s warning rephrased with teeth: you’re not going to see the failure in Q1. The world model will happily generate dashboards that show the KPIs moving in the right direction, because the world model is optimizing for the KPIs. What it doesn’t know is what the middle layer also knew — that the customer account you just flagged as “high expansion probability” actually hates your new CSM’s cold-email cadence, that the product line trending positive on unit economics is doing so because a sales rep is giving away features off-book, that the OKR you set for “reduce churn” is being hit by the CS team refusing to process cancellation requests. Middle management wasn’t just moving information around. It was editing it. It was catching the drift between what the dashboard said and what was actually happening, in real time, thousands of small corrections a day that never got logged because they were “just common sense.”

Stubhub is the visible version of this failure mode. A decade-plus ago they moved to all-in pricing — full transparency at checkout, no surprise fees at the end. Everyone in ticketing knew this was a bad idea. You compete on the headline price and collect the rest when the customer has already psychologically shown up at the concert. Stubhub’s all-in tickets looked 25% more expensive than the competition’s at the top of the funnel. Market share fell off a cliff. It took two or three years to claw it back. That was the good version — at least somebody at the top could see the revenue number moving, and eventually they reversed course. The malignant version of the same mistake is the one where revenue stays flat for six quarters because the world model is optimizing hard against the metrics you wrote down, and nobody notices the underlying rot until the annual plan comes in and the numbers don’t add up.

Anthony and I had this fight with an early Zynga guy years ago about OKRs. The system works, in theory. In practice, the sticky-by-design cadence — quarterly, not daily — is a feature when the middle layer is catching drift in real time between check-ins, and a catastrophe when the middle layer is gone. You don’t notice the drift. You notice the miss at the end of the quarter and you rewrite the OKR for next quarter, which also drifts. Compound interest works both ways.

What this means: If you’re the CEO of an enterprise, the lesson from Block is not “cut middle management faster.” The lesson is that the measurement layer underneath your world model has to be rebuilt before you start pulling people out. Every KPI in your org right now is a lossy compression of what someone was actually deciding. You manage what you measure. If the measurement is wrong, the world model will scale the wrong thing very efficiently for six months, and you’ll spend the next three years paying for it.

The X API Just Dropped 90%. Jevons Paradox Is About to Fire.

🦞 While the world-model argument plays out in slow motion across the Fortune 500, a smaller story landed on Saturday that’s going to have immediate consequences Monday morning. Robert Scoble DM’d Elon about the cost of the X API. Elon agreed to cut pricing from roughly $300 a day to roughly $30 a day, effective tomorrow. A 10x reduction.

Jevons Paradox, from a 19th-century English economist: when technology makes a resource cheaper to use, the total use of the resource goes up, not down. Jevons wrote about coal. The principle has shown up in every platform shift since — cheaper bandwidth created YouTube, cheaper storage created Dropbox, cheaper compute created OpenAI. Cheaper API access to X is about to create a category of AI apps that simply wasn’t economic at $300 a day.

The underrated primitive here is the list. Scoble built Aligned News by curating 50,000 AI accounts on X and running an AI over 40,000 posts a day. At $300 a day, that was one person’s operation. At $30 a day, every vertical gets one. A Claude-powered feed for luxury travel that watches TripAdvisor’s API and tells you which Caribbean hotels are trending among the right 500 people — that’s a business that couldn’t exist yesterday and can exist tomorrow. An aggregator for your board of directors that watches a hand-picked list of 300 competitors, customers, and industry voices — same thing. Taste becomes the moat. The person who can collate a 500-account list nobody else has is the new publisher.

CO/AI is going to subscribe to the X API directly this week — Monday or Tuesday, depending on our respective schedules — because we have accounts we weight higher than Scoble does and we want a direct look. At $30 a day, that’s a rounding error against what it buys you.

The signal: Don’t overthink what Jevons unlocks. The question to ask yourself tomorrow morning is not “should we build on the X API?” It’s “what’s the one list we already have — customers, suppliers, accounts, sources, competitors, researchers, investors — that becomes 10x more valuable the moment real-time social data costs a tenth of what it did yesterday?” If you don’t have a good answer to that, somebody else with a better list is about to be your new competitor.

What This Means For You

Two stories landed this week, and together they describe the shape of the next five years. A narrow class of founder-led AI-native companies is already operating at a cost structure the rest of the economy can’t touch. The rest of the economy is about to try to replicate it, and most of them are going to get the measurement layer wrong.

If you’re one of the three-person companies, know why it works. The edge isn’t the agents. The edge is that every person on your team can see every number and every workflow. The moment you hire a tenth person, the editorial function starts to distribute across people who weren’t in the room on day one, and your measurement layer has to be deliberate from that day forward. Don’t assume the magic scales.

If you’re running an enterprise, stop copying Fathom and start redesigning your metrics. Block cut 4,000 roles in February on a thesis that might be right and might be early. The difference between those two outcomes isn’t the quality of the AI. It’s whether you rebuilt the KPIs before you pulled the people out. If you can’t answer “which of these measurements would stop being a good measurement the moment an AI optimized for them,” you don’t have a world-model strategy. You have a slow-motion nail factory.

Subscribe to the X API this week. $30 a day. Not because you need it today, but because the cost of being wrong is now negligible, and the Jevons unlock happens for whoever builds first. Pick your list. Start watching it.

The companies that win the next five years aren’t the ones with the most agents or the fewest people. They’re the ones that figured out what to measure before they figured out what to automate.

Three Questions We Think You Should Be Asking Yourself

How do you know your dashboards capture what your middle managers actually decided? Pull the last ten meaningful decisions made by anyone at director level or below. How many of them are visible in a dashboard today? The gap between what gets tracked and what gets decided is where the world model will fail silently. If your answer is “I don’t know,” that’s your answer — and it’s the starting point for the rebuild.

Which of your current KPIs would become a nail factory the moment you rewarded an AI for optimizing against them? Run the exercise on your top ten metrics. For each one, ask what the maximum-gaming version of that metric looks like. Half of them will have a version that breaks the business. Those are the metrics you can’t automate against, or can’t automate against yet. The rest you can.

What’s the one list you already have that becomes 10x more valuable tomorrow? Customers you’ve known for a decade. Suppliers you’ve worked with since you founded. Competitors you’ve watched reshape their sector. Researchers you’ve followed on Twitter since 2019. Pick one. At $30 a day, you can now run a continuous signal over it. If you’re not the person building that tool, somebody with a worse list is about to.

“When a measure becomes a target, it ceases to be a good measure.”
— Goodhart’s Law, via Marilyn Strathern

We’ve been quoting this since 1975. The enterprise is about to learn why.

— Harry and Anthony

The Nail Factory

THE NUMBER: 4,000

You Can Build a Company With Three People Now. Here’s the Part Nobody’s Selling.

And Then There Is Block. And the 90% Behind It.

The X API Just Dropped 90%. Jevons Paradox Is About to Fire.

What This Means For You

Three Questions We Think You Should Be Asking Yourself

Sources

Past Briefings

Mind The Gap

Warp Speed, Fast, and Slow

Anthropic at $800 Billion. OpenAI Beaten by China. The Most Expensive Liquidation in History.