Skip to content
Fritz Gerald ZephirinMenu

Social

The Size of the Firm Is a Technology Variable

4 min read

In 1937, a 26-year-old economist asked a question so basic that nobody had bothered to answer it: if markets are so good at coordinating activity, why do firms exist at all? Why is the economy not just individuals contracting with each other for everything?

Ronald Coase’s answer earned a Nobel and still organizes the question better than anything since: using the market has costs — finding partners, negotiating, contracting, enforcing — and so does using an organization — managing, communicating, coordinating. A firm grows until coordinating one more activity inside costs more than buying it outside. The boundary of every company you’ve ever worked for sits, invisibly, where those two costs cross.

Which means firm size is not a fact of nature. It is a function of coordination technology — and it has moved every time the technology did. The telegraph and the railroad made continental corporations possible; the twentieth century’s giant integrated firm was an artifact of internal coordination being cheap relative to market transaction. Now run the Coase test on the current decade: AI just repriced both sides of the boundary at once. Inside the firm, cheap cognition collapses the cost of the connective tissue — reporting, summarizing, translating between teams, the middle layers that exist to move information (which is most of what org charts are). Outside it, agents and protocols collapse the cost of transacting — finding, specifying, verifying, standard interfaces for work itself. When both costs fall, theory alone won’t tell you whether firms get bigger or smaller. It tells you something more interesting: the equilibrium breaks, and size becomes a live design choice.

The evidence so far points both directions at once — honestly, and instructively. Toward smaller: revenue per employee at AI-era startups is rewriting the reference class — the fastest cohort in Stripe’s history, tiny teams reaching revenue that used to require departments, a solo-adjacent company selling for billions. Toward bigger: the same technology let the giants get more giant, and Meta paid $100 million packages to concentrate judgment in a smaller, denser core — while shedding the coordination layers around it. The resolution of the paradox: AI doesn’t shrink firms or grow them. It shrinks the coordination-shaped parts of every firm and lets the judgment-dense core set the size. Some cores are five people. Some are Meta.

I run a group of operating companies, which means I take the Coase test monthly, with real money. The pattern that has emerged in our own boundary decisions: what stays inside is whatever carries judgment, trust, or accountability — the definitions of resolved, the client relationships, the payroll promise, the compliance posture. What goes outside is whatever has become legible and verifiable — and AI is a legibility machine, constantly converting work from the first category to the second. The practical upshot for any founder: re-run the test annually, because the answer keeps changing under you. The function you had to own in 2022 may be a vendor in 2026; the boring layers especially.

Three honest caveats keep this from overclaiming. The IT era teaches humility — computers were supposed to flatten firms and instead concentrated them, because technology that lowers internal coordination costs fastest favors whoever coordinates most; AI could repeat that. Regulation sets boundaries that economics can’t move — banking and healthcare firms are the size compliance makes them. And the tiny-team data is young and survivorship-soaked; Stripe’s cohort is directional, not destiny.

But the direction of the question is settled even where the answer isn’t. For seventy years, “how big should this company be?” had a conventional answer: as big as the market opportunity, staffed to match. Now headcount and capability have decoupled — your effective size is your team times its tooling — and the honest planning question, the one we ask in every annual review, is Coase’s, updated: which of our coordination costs still justify employment, and which are now a protocol? Companies that keep answering from habit will carry the org chart of a technology era that ended. The ones that answer from the test get to choose their size — which is a stranger and more powerful freedom than it sounds.

Sources & further reading