The Coordination Floor

Five years ago, the numbers would have read as a typo.

A company reaches a billion-dollar valuation with seven employees on the payroll. Another crosses the line with ten, another with twenty-nine. A venture dataset that tracked nearly a thousand unicorns from 2010 onward turned up a small cluster — twenty-six companies, most founded in 2022 or later — that hit a billion dollars with a median of thirty-four people and a median value of four billion. The wider unicorn population needed a median of four hundred employees to cross it.

The instinct is to file this under AI hype and move on. That would be a mistake. The pattern is measurable, and almost everyone looking at it is reading it wrong. What follows is an attempt to get at what the number actually means — the mechanism underneath, rather than either the celebration or the dismissal it usually draws. Name that mechanism and the same lens that explains why these companies can be so small will also show you where the model stops working.

Two readings, both incomplete

There are two dominant ways to interpret the tiny-team billion-dollar company, and they are both partly right.

The first is the popular reading: headcount is obsolete. AI has made the employee optional; the one-person billion-dollar company is imminent; anyone with a laptop and a subscription can now do what once took a department. It is an exciting story and it captures something true — the tools are extraordinary — but as an explanation it collapses on contact with the data. The companies in question are not one person. They cluster around thirty. They are small, not absent. Something is still holding them at a size, and “headcount is dead” cannot tell you what.

The second is the investor’s reading: this is talent density, priced on capability. A handful of world-class researchers from a few frontier labs command enormous valuations because the market is betting on what they might build, and because denying that talent to a rival is itself worth billions. This is more rigorous, and closer to the truth. But it accounts for who these people are and why investors pay, without touching the question of why the structure itself works. It explains the price and leaves the shape unexplained.

Both readings answer how these teams got so valuable and skip the harder question: why they are so small — and why the old ones were so large.

Naming the variable

Start with a fact so ordinary it is easy to miss: adding a person to an organization does not just add their work. It adds their coordination — the meetings they attend, the approvals they wait on, the context they must be given, the decisions that now route through one more node. The output of a team scales roughly with its size. The cost of keeping that team aligned scales faster. Every additional person adds a line of work and several lines of communication.

This is the tax that has always sat underneath the org chart. For most of business history it was invisible because it was unavoidable — a fixed cost of doing anything ambitious. If you wanted to build something that took a hundred people’s worth of work, you hired a hundred people, and you paid the coordination cost that came with them: the managers to run them, the layers to route decisions, the process to keep the layers honest. Nobody experienced this as a tax. It was simply how companies got built.

Here is the reframe the whole phenomenon turns on. The old four-hundred-person unicorn was not four hundred people’s worth of value creation. It was some smaller amount of value creation, wrapped in the coordination structure required to sustain it at that scale. Much of that headcount existed not to make the product but to manage the fact that so many people were making it. The team was large partly because the team was large.

This site has a name for that hidden charge already: the coordination tax — the cost, compounding with every element added to a system that was never reoriented, of holding the thing together. What the micro-unicorn data exposes is the tax’s lower boundary. Call it the coordination floor: the minimum organizational size below which the old way of building could not produce a billion dollars of value, because the tools of the era required a department to wield them. Building a model took an infrastructure team. Reaching customers took a sales organization. Running the systems took operations. Each capability came bundled with the people to operate it, and those people came bundled with the cost of coordinating them.

What changed in 2022 is that the floor dropped. The AI-era stack lets a very small team wield capabilities that previously demanded a department — not because people became unnecessary, but because the department did. When one capability after another collapses onto a laptop, the team no longer has to grow to reach them — and if it doesn’t have to grow, it doesn’t have to pay the coordination tax that growth imposes. What you get is not a company with no people, but a company that stays deliberately below the floor and keeps for itself the leverage that used to go into holding a large organization together.

Through the coordination lens the practice reads as a deliberate expense: a trillion-dollar company paying, twice a week, to hold the coordination tax down — keeping information shared and decisions fast at a size where the default runs the other way. The startups get that condition for free by staying small; Nvidia has to manufacture it against the grain of its own size.

Coordination cost — not talent, not capital, not headcount — is the binding constraint on how fast an organization turns intelligence into value. Or rather it was: as the economics of building shifted from headcount to compute, the tax did not get cheaper so much as stop being the thing that binds. AI left the team intact and lifted the floor that used to force it to grow.

The mechanism is old. Only the escape is new

None of this began in 2022. The coordination tax has a literature, and it is fifty years deep. In The Mythical Man-Month, Fred Brooks — who had just run IBM’s OS/360, one of the largest software efforts of its era — observed that the communication paths in a team grow as n(n−1)/2. Ten people carry forty-five channels; fifty people carry more than twelve hundred.4 The work rises linearly with headcount; the cost of keeping the work aligned rises with the square of it. That curve is the shape of the floor.

It has since been measured directly. A large-scale study of fifty-eight open-source projects — more than thirty thousand developers and half a million commits — found that the output of the individual developer falls as the team grows, a software-era confirmation of the Ringelmann effect first seen in nineteenth-century rope-pulling experiments.5 Separate analyses put the most productive team size at three to seven people, and find per-developer commit rates collapsing on projects past fifty contributors.6 Across three datasets and three eras the finding holds: coordination cost is super-linear, and it has always set a ceiling on how much a growing team could actually get done.

This is the same mechanism the essays on this site keep meeting from different directions. After Brooks already put the case that the scaffolding built to manage the human coordination tax is no longer load-bearing, and that the ceiling Brooks described is gone; the micro-unicorn cohort is what the far side of that removed ceiling looks like with numbers attached. Coordination is essential complexity, the part no tool removes — the micro-unicorn simply shed enough of the accidental kind to stay under the ceiling. It is also why fixing one bottleneck rarely helps: the constraint just moves down the line, the behavior the “tax is conserved” reading below describes. AI did not repeal the law; it changed which side of it a small team can live on.

The evidence, and its limits

The clearest numbers come from a mid-2026 analysis that pulled roughly 961 unicorns from standard venture databases and isolated the companies that reached a billion dollars with fewer than a hundred people within two years of founding — twenty-six of them in the post-2022 cohort. Against the full population, the contrast is a category change: the median unicorn generates about five and a half million dollars of value per employee; the median micro-unicorn, a hundred and eighteen million.1

Intellectual honesty requires stating just as clearly what these numbers are not — because the caveats are where a careless version of this argument falls apart.

None of this weakens the core finding; it sharpens it. The leverage is real and it is bounded, and one theory accounts for both — the same reasoning that explains the leverage marks where the bound sits. The macro data runs in the same direction: in a 2026 survey of more than ten thousand executives, 88% were experimenting with AI while 81% reported no meaningful bottom-line impact — the gap sitting not in the tools but in whether the organization around them was redesigned to use them.8

Where the mechanism predicts the break

A good explanation earns its keep by telling you when it stops applying. The coordination-cost lens does this cleanly. The three most serious risks facing these companies — each named by the same analysis that celebrated them — turn out to be one thing wearing three faces: the coordination tax reasserting itself.1

One: the missing layer is not waste. The sharpest critique of the tiny team is that it underwrites capability but cannot deliver execution at enterprise scale. What a twenty-person company definitionally lacks is the middle layer — the people who own accounts, manage delivery, handle escalations, translate a research team’s output into a product an enterprise can depend on. It is tempting to call that layer bureaucracy. The mechanism says otherwise: that layer is coordination made visible — the necessary cost of serving many customers reliably over time. A team stays below the floor only as long as its value comes from capability rather than sustained, at-scale delivery. The moment durable enterprise revenue becomes the goal, the floor comes back, because reliability at scale is a coordination problem. The tax was never abolished. It was deferred.

Two: leverage per person is fragility per person. When a company’s worth is concentrated in a handful of people, losing any one of them moves a meaningful fraction of that worth out the door — and in this cohort, entire founding teams have walked to larger rivals. High leverage and high fragility are the same fact seen twice. A large organization is robust to any individual departure precisely because it has diffused its knowledge across a coordinating structure — the very structure the lean team shed to move fast. You cannot keep the resilience of the large org and the speed of the small one; they are traded along the same axis.

Three: the circular market. In this cohort, the dominant investor and the dominant supplier of the compute these companies run on are frequently the same party — capital flows in and returns as spend on the supplier’s chips, a loop that can inflate demand in a self-reinforcing circle and echoes vendor-financing dynamics seen before past corrections. This is a coordination cost that moved up a level, from inside the firm to inside the ecosystem. The company externalized its coordination by staying small; the system absorbed it, and concentrated the risk.

The coordination cost a small team avoids does not vanish; it relocates. The tax is conserved — you can choose where to pay it, not whether.

What it means, for whoever is reading

For the founder or operator, the point is not to stay tiny forever. It is that growth is a decision with a price, and the price is coordination. The instinct to hire against every problem re-imposes the exact floor that AI just lifted. The discipline worth keeping is knowing which additions buy genuine capability and which merely buy the need to coordinate the additions.

For the incumbent, the micro-unicorn is not a threat to imitate by getting small — most cannot — but a mirror. The speed those tiny teams have for free is the speed large organizations pay dearly to reconstruct, whether through two-pizza teams, autonomous product units, or a chief executive running an agenda-free room twice a week. The work is not to shrink the company but to hold the coordination tax down as it grows — which is why the firms that redesign around AI, rather than bolting it onto the existing structure, pull away: BCG put a 3.6× shareholder-return gap between the two, and it is the organization, not the technology, that was always the constraint.7

For the board or allocator, the useful question is diagnostic: is this company’s leanness capability that hasn’t yet met its coordination bill, or a durable operating advantage? The first is a stage; the second is a moat, and the value-per-employee number cannot tell them apart. The mechanism can, by asking whether the company’s value still comes from what it could build or from what it reliably delivers at scale.

Coda

The micro-unicorn does not show that people stopped mattering. It shows that coordination was always the constraint — the hidden floor under the org chart, the reason a billion dollars of value used to require four hundred people to hold together. AI lowered that floor for a particular kind of work, for now, without abolishing it. The teams thriving below it are not evidence that the tax is gone; they are a measure of how large it always was, made visible by the first tools capable of not paying it.

Whether the floor stays low, and for how many kinds of company, is the open question — and the one worth watching more closely than the headcount figures that keep making headlines. The coordination tax is one dimension of a larger reading: the Human judgment that decides what to build, the Organization structure that couples it to the work, and the Technology that carries it — designed as one machine rather than a lean team that will one day rediscover, the hard way, the cost it thought it had escaped. That is the move from human-centric to human-agentic the rest of this canon works through, and the reason the answer is to build the agentic organization alongside the legacy one and let it prove itself rather than to shrink. The framework is HOT; moving a real firm through it rests on a named discipline rather than good intentions — Salim Ismail’s ExO 3.0, the spine for redesigning an organization around intelligence instead of retrofitting it onto one.

Related on this site: After Brooks · The Exposure Curve · Custom Is Making a Comeback · One Move, Three Faces · The Problem Was Never the Technology. The framework in full: Human · Org · Tech.