Coase's Theory of the Firm

The coordination-cost ceiling on firm size

The economist Ronald Coase argued that a company grows only until the internal cost of communication and coordination becomes too great to justify, which is what caps the size of firms. In an interview with Garry Tan, Ryan Petersen uses the idea to frame the central scaling question: if coordination cost is what limits firm size, then anything that lowers that cost should raise the natural size a firm can reach.¹ Tan, in the same conversation, suggests that "the power of software" may be doing exactly that, connecting it to the idea of many product lines living under one roof, made viable by decentralization plus the right rules between units.

Petersen frames his own team-of-teams design at Flexport as an attempt to beat the Coasean limit. If the firm is a set of small autonomous teams that do not need to coordinate constantly, the coordination-cost ceiling rises and the company can grow larger before the cost overwhelms the benefit. His caveat, in the same account, reaffirms Coase's core insight rather than dispensing with it: without a shared data model and APIs to keep the teams connected, there is no firm at all, and, as he puts it, "you might as well spin out 26 separate companies." Coordination cost can be lowered, but it cannot be ignored.

Lowering the cost versus removing it

The theory functions as the backdrop for the broader org-design discussion, in which specific mechanisms are read as different attacks on the same coordination cost. The API-mandate discipline of routing team-to-team communication through clean interfaces, and Petersen's small autonomous teams, both create more efficient relay channels, lowering coordination cost without eliminating it.

Jack Dorsey has advanced a stronger claim in the same frame. His argument, applied to his transformation of Block, is not that AI lowers the cost of information relay across many people but that it removes that specific cost. His reasoning is that every artifact a remote-first company produces is already structured information, so an intelligence layer ingesting all artifacts makes them queryable directly, and a CEO no longer needs a chain of managers to synthesize the synthesis of a team's output. In that telling the relay function of the hierarchy is replaced, and only the judgment, accountability, and coaching functions remain, which is why Dorsey describes flattening Block from a pre-transformation depth of five toward a target of two or three and an ideal of one, all 6,000 employees mediated through the intelligence layer and reporting to him.² The distinction he draws is that efficient relay channels lower the coordination cost while an intelligence layer aims to remove the relay entirely.

What stays unresolved

Whether software or AI permanently changes the answer is presented as an open question rather than a settled result. One line of doubt is whether software changes the natural size of firms or merely defers the limit. Another is whether an intelligence-layer model eliminates the Coasean ceiling or simply relocates it: if the quality of the shared world model becomes the bottleneck, the coordination cost has not disappeared so much as concentrated in model curation and alignment, and the ceiling re-emerges at a new layer. The same reasoning has also been extended in the opposite direction, toward coordination cost dropping not just for the firm but down to the level of the individual task, which would dissolve the firm's lower boundary rather than raise its upper one. In each case the underlying Coasean question is the same, and only the proposed mechanism differs.

The theory connects directly to desired state org design, which attacks coordination cost by making decisions about org structure computable rather than negotiated, and to single-threaded teams as a concrete form of the small, loosely coupled unit that raises the ceiling.