Why AI Agents Don't Dissolve the Firm

In 2012 Facebook paid roughly one billion dollars for Instagram, a company with thirteen employees at the time, per contemporaneous press coverage of the deal. In 2014 it paid nineteen billion for WhatsApp, then around fifty employees, again per the widely-reported deal terms. Both figures get repeated as proof that software let a company become enormous without becoming large — that headcount had already stopped being the ceiling on value, years before anyone could type a prompt into anything.

AI agents force the sharper version of the same question. If the ceiling was lifting before a model could write code, answer a support ticket, or clean a customer list, what happens to a firm once the labor itself can be discovered, priced, instructed, and dismissed inside a single API call?

This site's name for what sits on the other side of that question is the Zero-Employee Organization (ZEO): a firm designed so its output stops scaling with its headcount. But the theory behind that name is not new, and saying so plainly is most of the argument here, not a footnote to it. The short version, before the theory that gets you there: the firm doesn't dissolve. It concentrates — the same standing authority, held by fewer people, sometimes just one.

A ninety-year-old question, asked before anyone could type "AI"

In 1937 the economist Ronald Coase published a short paper called "The Nature of the Firm." His question had nothing to do with computers: if prices coordinate a market economy so efficiently, why does a firm exist at all — a little island of command where a boss just tells people what to do, instead of everyone re-negotiating a fresh contract for every task?

Coase's answer was that using the market is not free. Every time you buy something instead of making it yourself, you pay a toll: finding a supplier, agreeing a price, writing a contract, watching for someone cutting corners. Bundle enough of those tolls together and it becomes cheaper to hire someone under a standing employment contract — an open-ended agreement to take direction — than to write a new contract every time a job needs doing. A firm, in this view, isn't a natural unit of production. It's a cost-minimizing alternative to constant market negotiation, and it stops growing exactly where running one more task in-house costs more than buying it outside.

That's the whole mechanism an AI agent seems to attack directly. An agent can be found, briefed, supervised, and let go inside one conversation, with no onboarding and a complete log of every instruction it was given. If Coase's story is really a story about the cost of coordinating through the market, then coordinating through an AI-saturated market has never been closer to free. So does the firm dissolve into a cloud of on-demand agentic contracts — or does it thicken around the one thing that still can't be procured through an API: a human who holds the equity, bears the liability, and is accountable when something goes wrong?

Coase's answer was incomplete on purpose — and that's where the theory actually gets useful

Coase named the cost of the market. He didn't break it down, and economists spent the next four decades doing that. The most load-bearing refinement came from Oliver Williamson, across a 1979 paper and a 1985 book, who argued that whether a task belongs on the market or inside a firm turns on three things happening together:

Put those three together and you get the hold-up problem: once you've sunk a relationship-specific investment, your counterparty can renegotiate the terms in their favor, because you can't walk away without destroying the value of what you already built. Rational parties see this coming and either underinvest — or pull the whole transaction inside one firm, where a single authority resolves disputes by decision instead of by fresh negotiation. That is Williamson's account of what a firm actually buys: not just cheaper contracting, but the right to settle an argument by fiat instead of by a court or a renegotiated deal.

A later refinement, from Sandy Grossman and Oliver Hart in 1986, sharpens the question further: what does bringing a transaction inside a firm actually transfer? Their answer is residual control — the right to decide anything a contract doesn't cover. Ownership, in this framing, is not a feeling of control; it's the specific right to make the call when the paperwork runs out.

What cheaper coordination actually touches, and what it doesn't

An AI agent attacks the first of Williamson's three legs hard: bounded rationality, in the narrow sense of the cost of communicating instructions and monitoring compliance. It does very little, on its own, about the other two.

Cheap coordination doesn't dissolve hold-up risk. It relocates it — from the factory floor and the supplier contract to the system prompt and the standing permissions an agent was granted once and never revisited.

The reframe: not dissolving, concentrating

A large share of commentary about AI and the firm quietly equates "the company has one employee" with "the firm has dissolved." Coase's own definition says otherwise. The firm is the substitution of standing authority for repeated market contracting — and a solo founder directing a fleet of agents by continuous instruction, without renegotiating a fresh contract for every task, hasn't dissolved that relation. She has shrunk its human membership to one while keeping the relation itself fully intact, and in fact concentrated more of it into fewer hands than a management hierarchy ever held at once.

Instagram's thirteen employees and WhatsApp's fifty didn't mean either company had stopped being a firm in Coase's sense. Both still needed a legible ownership structure, human counterparties for payments and advertising infrastructure, and — eventually — total absorption into a much larger firm with the headcount to handle what a thin founding team couldn't. The firm didn't disappear. It delegated everything except product vision and equity to somebody else's larger firm, which is a make-or-buy decision at the scale of the whole company, not evidence the boundary vanished.

And the boundary doesn't always move for cost reasons at all. Uber and similar platforms used app-based coordination to strip out most of the search and monitoring cost that once justified direct employment — exactly what Coase and Williamson would predict pushes a transaction toward the market end of the spectrum. California's Dynamex decision (2018) and AB5 (2019) pulled it back toward employment by law; the platforms then spent north of $200 million, per widely-reported campaign spending disclosures, on a ballot measure, Proposition 22 (2020), to carve drivers back out again. The firm boundary here was redrawn by courts, a legislature, and a ballot initiative — a reminder that "make or buy" is a legal and political question as much as an economic one, and that falling coordination costs don't mechanically translate into a redrawn boundary if regulators insist otherwise.

What this theory is not claiming credit for

None of the above is new theory wearing a new coat of paint. Coase, Williamson, and Grossman-Hart-Moore did the actual work, across five decades, well before anyone was training a language model. What is genuinely new is the case now sitting in front of the theory: labor that can be procured, instructed, and released at a speed and price the classic literature never had to model. Calling that a Zero-Employee Organization is a name for the case, not a claim on the ideas underneath it — and the honest version of this argument credits the lineage every time, rather than letting a live-sounding acronym imply the thinking started here.

And the same caution the GM/Fisher Body callout raises against a confident real-time story applies to this piece's own "concentrates, not dissolves" verdict, not just to other people's takes. That reframe is this essay's best guess at how a ninety-year-old theory maps onto a case still unfolding, argued from the theory's own logic rather than from data that doesn't exist yet — it is not a settled finding the way the theory's founding papers are. The GM story is the reminder to hold that distinction, applied to this piece and not only to the ones it's arguing against.

This piece leans entirely on economic theory dated 1937 to 1990 and public case law, so it carries no rot register — nothing here is a claim about what a specific AI product can currently do, and the underlying argument doesn't expire on a product release cycle.