the minimum viable firm

the cost of serious work collapsed, the team you need shrank to four, and the thing that forms those teams still does not exist

11-Jun-26

The cost of producing a unit of cognitive work fell by one to two orders of magnitude in about two years. A financial model, a research memo, a design draft: the price of making one collapsed. That is a change in cost structure you can see in what people now pay for the output.

Here is what follows. When a serious deliverable needed a team of twelve, you needed an institution to hold them: payroll, offices, management layers, the overhead of keeping all those people pointed the same way. When four people with disciplined use of AI produce the same output, the institution becomes optional. The minimum viable team for serious professional services dropped from roughly fifteen people to roughly four.

That is the whole thesis. Everything else follows from it.

the glass wall

In 1873 a zoologist dropped a glass partition between a pike and the minnows it ate. The pike charged, hit the glass, charged again, and eventually stopped trying. Then he lifted the glass out. The minnows swam past its open mouth and the pike starved, still convinced the wall was there.

Most professionals are the pike. The wall was the coordination cost of large organizations, the reason you needed a big company to do serious work. AI pulled the wall out. The clients, the revenue, the meaningful work are swimming right past, and most people still flinch at glass that is gone.

where the value moves

Execution is becoming abundant. Judgment is becoming scarce. Every competitor has the same tools now, which makes the crowded markets bloodier and pushes the real edge somewhere the tools cannot reach.

Code production gives way to deciding what to build. Content generation gives way to knowing what matters. Financial modeling gives way to choosing the assumptions. Research synthesis gives way to holding a worldview steady under noise.

AI fluency itself is a depreciating asset. The gap between power users and everyone else narrows every six to twelve months. What compounds is domain expertise, relational trust, and the judgment of a team that has learned to work together under pressure. Those appreciate while the tool-skill expires.

the infrastructure gap

The “leave corporate and build your own thing” genre has produced real insight. It has also become its own crowded market. The people who write best about post-corporate work diagnose the infrastructure gap correctly, then sell content that sits on the near side of it. Their revenue runs on a steady supply of aspiring founders, so graduating people efficiently into working firms cuts against their own model.

The market is flooded with recipes. The kitchen is still missing.

A kitchen would hold the parts a working firm needs before it has revenue. Partner matching on role complementarity and aligned income floors, rather than networking events and personality quizzes. Governance designed while everyone is still friendly: operating agreement, vesting, exit protocol, decision rights, written before the first real fight. A shared operating stack installed early: CRM, decision log, financial dashboard, the AI workflows. A trust test built on observed behavior with real money at stake, instead of self-report. Distribution into one specific scene united by a worldview. And failure protocols that allow a clean dissolution while exit is still cheap.

what we don’t know yet

The honest version of this idea names what it does not know. The base rate of micro-firm survival is unmeasured, and the answer changes everything: forty percent at five years is a genuine alternative to a job, fifteen percent is a high-risk bet. The real AI productivity multiplier is unsettled too. The model wants something like ten to fifteen times; if the lived number is three to five, the economics shift hard. And the absence of formation infrastructure proves only that nobody built it, never that anyone wants it.

The only way to answer those is to run one cohort and watch. The thesis needs no macro forecast to hold. Whether the broad disruption lands in 2030 or 2040, the structural economics of small-team professional services have already moved. One real cohort would be worth more than another year of research, because it would produce the first honest failure data in a field drowning in survivorship bias.

someone should build this

The infrastructure that turns an experienced refugee from a big firm into a functioning small one does not exist yet. Build it as a staged process with a financial commitment at every gate, so the wrong people leave early and cheaply. Intake and matching first. Then a paid trial on a real client at a discount, with the revenue held back, because watching people work under mild money pressure is the only trust test that tells the truth. Then governance, designed while the room is still cooperative. Then launch: one offer, one scene, the stack switched on. Four people, twelve weeks, and you walk out with a firm.

Start with one cohort. If it works, scale it. If it breaks, you still produced something the whole category lacks.

Someone should build this. I keep thinking it might be me.