The Cognitive Organisation: Coase and the Cost of AI

In 1937, a young British economist named Ronald Coase asked a disarmingly simple question: if free markets are the most efficient mechanism for allocating resources, why do business firms exist at all?

His answer founded the economics of organisation. Markets are brilliant, Coase noted, but using them is expensive. Discovering prices, negotiating terms, drawing up bespoke contracts, and inspecting deliverables all incur “transaction costs.” Where those costs are high—particularly for complex, uncertain, or highly customised work—it is cheaper to bring people under one roof and coordinate them through managerial command.

The firm is essentially an island of administrative planning in a sea of market prices. Hierarchy exists because talk is expensive.

The price of alignment

For nearly a century, management theory treated this trade-off as an iron law. Routine goods were bought on the open market via standardised contracts; complex knowledge work was internalised inside corporate silos.

Middle management became the human routing switch for enterprise data. Their primary economic function was “semantic alignment”: translating board directives into division goals, division goals into project specifications, and project specifications into daily tasks, all while translating customer feedback back up the hierarchy.

This translation machinery was notoriously sluggish and expensive, but firms tolerated it because the market alternative—drafting legal contracts for every internal conversation—was unthinkable.

The collapse of transaction costs

Generative artificial intelligence directly attacks the economic rationale for hierarchy.

An LLM is not merely an automated copywriter; it is an engine for instantaneous semantic alignment. It translates intent between natural language, legacy database queries, structured code, and proprietary APIs at marginal costs approaching zero.

Consider the implications:

1. The dissolution of the internal interface. The traditional rationale for enterprise ERP systems was to force every department into a single, uniform data model. If AI agents can translate idiosyncratic local workflows into systemic compliance on the fly, the need for rigid centralisation evaporates.
2. The return of the bazaar. If contract generation, code synthesis, and verification become frictionless, the boundary of the firm shifts inward. Projects that previously required a permanent team of eighty employees can now be executed by a five-person syndicate coordinating external models and automated agents.
3. The crisis of middle management. If routing, summarising, and aligning information is automated, the managerial tier whose job was to insulate executives from frontline operational detail becomes an expensive bottleneck.

Beyond the corporate boundary

Open-source software communities provided the dress rehearsal for this shift. For decades, thousands of autonomous developers have coordinated globally complex operating systems without employment contracts or middle managers, relying on modular codebases and peer review.

What open source lacked was cheap, instantaneous translation across non-technical domains. AI supplies precisely that missing link.

Firms will not disappear overnight, but their historical justification has been severely undermined. When semantic alignment is no longer a scarce good, the corporate pyramid begins to look less like an efficient engine of enterprise and more like an expensive legacy monument to obsolete communication costs.