When AI makes cognitive capability free, trust becomes the only scarce input. That shifts the organizing logic of economic institutions — from Coase's transaction costs to Dunbar's neurological ceiling on human relationship capacity. The firms that survive aren't the most capable. They're the most trusted.
The wrong question
Every post-labor economics conversation starts with the same framing: what do humans do when machines do everything? The answers are predictable. Universal basic income. Creative pursuits. Leisure. Retraining. New jobs we can't imagine yet.
All of these assume the interesting question is about activity — what occupies human time. None of them ask the structural question: what forms of economic organization emerge when cognitive capability is no longer scarce?
That's the question that matters. Not what individuals do. What institutions become.
Coase in the age of agents
Ronald Coase won the Nobel Prize for answering a deceptively simple question: why do firms exist? If the market is efficient, why don't individuals just contract with each other for every task? His answer: transaction costs. It's expensive to find the right person, negotiate terms, monitor performance, and enforce agreements for every individual task. Firms exist because bundling these transactions inside an organization is cheaper than executing them on the open market.
Every expansion of the firm — every new hire, every department, every layer of management — is a response to the same calculation. Internal coordination costs less than external contracting. The firm grows until the marginal cost of internal coordination equals the marginal cost of market transactions.
AI agents invert this. When an agent can handle tasks that previously required a human employee — research, analysis, writing, code, scheduling, monitoring — the transaction cost of getting work done drops toward zero. But only for capability. The transaction cost of trust doesn't change.
This is Coase's insight applied to cognitive hyperabundance: the firm shrinks to the minimum viable trust network, because everything that used to require headcount for capability reasons can now be handled by agents. What remains is the irreducibly human part — the judgment calls, the relationship management, the contextual understanding, the alignment on values and direction.
Dunbar's constraint as economic structure
Robin Dunbar's number — roughly 150 — describes the maximum number of stable social relationships a human can maintain. The constraint isn't time or technology. It's cognitive. You can only model about 150 people well enough to predict their behavior, trust their judgment, and understand their context.
In a capability-scarce economy, this constraint was irrelevant to business structure. You needed 500 people to run the operation, so you built hierarchies, processes, management layers — all designed to coordinate people who didn't need to trust each other deeply. The org chart is a trust-substitution mechanism. It replaces genuine understanding with role-based authority.
In a capability-abundant economy, Dunbar's number becomes the structural constraint on the firm. You don't need 500 people. You need the people you trust, plus agents. The org chart collapses because the trust-substitution mechanism isn't necessary when you only have 5 people and they all understand each other's context.
The prediction: economic activity reorganizes around trust networks of 2-50 people, each amplified by orders of magnitude through agentic infrastructure. Not because small is ideologically preferable. Because trust doesn't scale past Dunbar's limit and capability no longer requires headcount.
The fractal pattern
This scales fractally across every organizational relationship.
Founding. The old model: find co-founders with complementary skills. I can't code, you can't sell, we partner. That's a capability gap argument. AI closes every capability gap. The only remaining reason to found something together is alignment — shared values, shared judgment, shared trust. Skills are abundant. Alignment is scarce.
Employment. The old model: hire people who can do the work. The new model: the work is done by agents. Hire people whose judgment you trust to direct the agents. The hiring criterion shifts from "can you do X" to "do I trust your taste about what X should be."
Client relationships. The old model: clients buy capability. The agency that's better at design wins the design contract. The new model: capability is commoditized. Clients buy trust — the confidence that someone understands their situation deeply enough to architect the right solution. The org-to-client relationship mirrors the co-founder relationship: small, high-context, alignment-first.
Markets. The expansion era rewarded scale — biggest platform, largest audience, most reach. The trust contraction rewards depth — deepest understanding, thickest context, most durable relationships. Global reach persists but the actual unit of economic organization shrinks.
What the contraction looks like
The trust contraction doesn't mean everyone becomes a freelancer. It means the shape of organizations changes.
Large firms don't disappear. They hollow out. The core shrinks to the trust network — the people who understand the business deeply enough to direct its AI infrastructure. Everything else becomes either automated or contracted to smaller trust networks that specialize.
New firms form around trust, not capability. Two people who trust each other's judgment completely, plus agents, can do the work of a 50-person company. The formation event isn't "we found a market opportunity" — it's "we found alignment." The market opportunity is downstream of the trust.
Professional services — consulting, law, finance, design — restructure around depth rather than breadth. The firm that serves 500 clients superficially loses to the firm that serves 30 clients with such deep contextual understanding that the AI infrastructure it builds actually works. Meaning architecture becomes the core competency: the ability to translate a client's situation into context rich enough for agents to operate on.
Geographic and cultural clustering intensifies. Trust forms more easily with shared context — shared language, shared cultural norms, shared professional background. The trust contraction pushes economic organization back toward locality even as technology enables global reach. You can serve clients anywhere. You form trust networks locally.
Where this leads
The trust contraction is not a temporary market correction. It's a phase transition in the structure of economic organization, driven by the commoditization of cognitive capability and bounded by neurological constraints on human trust formation.
The resulting organizational forms are predictable from first principles — small, high-context, trust-dense networks with agent-powered leverage. Coase explained why firms exist. Dunbar explains where they stop scaling. Granovetter, Ostrom, and Putnam each mapped a piece of how trust organizes economic action. None of them anticipated a world where capability is free and trust is the only scarce input.
The empirical evidence is early but directional. The Anthropic-Pentagon data — where the market selected for the trustworthy model over the compliant one — is a natural experiment in trust-as-selection-pressure. The formation patterns of AI-native startups — tiny founding teams, massive agent leverage — are organizational data points. The question that ties them together: In an economy where cognitive capability approaches zero marginal cost, what organizational structures emerge when trust is the binding constraint on coordination?
Sources
- Coase, Ronald — "The Nature of the Firm" — Economica — 1937
- Williamson, Oliver — "The Economic Institutions of Capitalism" — 1985
- Dunbar, Robin — "Neocortex Size as a Constraint on Group Size in Primates" — Journal of Human Evolution — 1992
- Granovetter, Mark — "Economic Action and Social Structure: The Problem of Embeddedness" — American Journal of Sociology — 1985
- Ostrom, Elinor — "Governing the Commons" — 1990
- Putnam, Robert — "Bowling Alone" — 2000
- Anthropic — Market data from Pentagon standoff period (Feb-Mar 2026)
- Ramp Procurement Data — Enterprise adoption shifts (2025-2026)