Coase's insight reframed the firm from a producer of goods to a coordinator of activities. The firm exists because internal coordination — hiring employees, directing their work through managerial authority, monitoring compliance — is often cheaper than repeated market transactions with independent contractors. This coordination includes both production coordination (who does what, when, how) and direction coordination (what should be done at all, for whom, to what standard). AI unbundles these two functions. Production coordination — scheduling, routing information, tracking progress — is precisely the kind of mechanical, rule-governed activity AI performs well. Direction coordination — exercising judgment about priorities, quality, and strategic direction — remains irreducibly human. The firm of the AI age exists primarily to house the second function.
Chester Barnard's 1938 Functions of the Executive offered a complementary framework: organizations exist not merely to economize on transaction costs but to create the cooperative conditions that productive work requires. Barnard identified three functions executives perform: maintaining communication channels, securing essential services from organizational members, and formulating purpose. AI can absorb the first (information routing), partially automate the second (monitoring and incentive design), but cannot perform the third. Purpose formulation — deciding what the organization is for, what it should build, whom it should serve — requires human commitment and accountability that computational systems do not possess.
Mary Parker Follett distinguished in the 1920s between power-over (hierarchical command) and power-with (collaborative capability generation). Traditional firms relied heavily on power-over coordination: managers directed, workers executed. AI-augmented organizations must rely more on power-with: small groups exercising collective judgment, with AI handling execution. The shift from command to collaboration changes what coordination means — from the routing of tasks to the alignment of understanding, from monitoring compliance to cultivating shared purpose. These are social functions that cannot be automated because they depend on trust, shared context, and mutual accountability rather than on information processing.
The concept that firms exist primarily for coordination rather than production emerged gradually from Coase's 1937 insight. Coase himself focused on transaction costs as the explanation for firm boundaries. The coordination framing became explicit in the 1970s work of Oliver Williamson, who argued that firms are governance structures designed to manage transactions that markets handle poorly — particularly transactions involving asset-specific investments, uncertainty, and information asymmetry. The AI moment makes the coordination function analytically distinct from the production function in ways the earlier literature did not anticipate, because AI can now perform much of what firms historically did while leaving coordination needs unmet.
Production versus direction coordination. The firm historically bundled two functions that AI now unbundles: managing who does what (increasingly automatable) and deciding what should be done (still human).
Coordination as social infrastructure. The irreducible coordination functions are trust-building, tacit knowledge transmission, standard-maintenance, and professional belonging — none of which markets provide efficiently.
The managerial unbundling. The manager's role divides into information routing (AI-absorbable) and leadership (setting standards, developing talent, owning painful trade-offs in public) — only the second justifies organizational hierarchy.
The vector pod as pure coordination. Small cross-functional groups that decide what to build without building it represent the Coasian firm stripped to its coordination essence — judgment without production overhead.