The Firm in 2026

Return to Coase

In the first post of this series, we asked a question that seemed almost quaint.

Why do corporations exist?

The answer — because markets have transaction costs — was not new. Coase established it in 1937. Williamson refined it. North embedded it in institutional theory. The framework was clear: firms internalise activities when the cost of coordinating through markets exceeds the cost of coordinating through hierarchy.

We decomposed those costs into six components:

1. Search costs
2. Matching costs
3. Contracting costs
4. Monitoring costs
5. Enforcement costs
6. Managerial bandwidth costs

And we stated the boundary condition:

TC_market = TC_internal

The firm exists where internal coordination is cheaper than market coordination. The boundary shifts when any cost component changes.

Fourteen posts later, every cost component has changed.

This post closes the loop.

The Ledger, Revisited

We now have enough machinery — protocols, agent networks, value streams, network geometry — to return to each transaction cost and assess what has actually happened.

Not in theory. In structure.

1. Search Costs — Collapsed

Search was the first cost to fall.

In 1937, finding a supplier meant trade directories, word of mouth, physical proximity. Search was slow, expensive, and geographically bounded.

By 2010, platforms had reduced search costs by orders of magnitude. Google, LinkedIn, Upwork — each made discovery nearly instantaneous across global pools.

By 2026, AI agents have pushed this further. Agents do not search passively. They actively scan, filter, rank, and shortlist. A sourcing agent does not wait for a job board to surface candidates — it crawls structured and unstructured data, evaluates fit against multi-dimensional criteria, and presents ranked options in seconds.

Search cost has not merely fallen. It has been restructured. What was once a human activity — browsing, reading, comparing — is now an executable workflow.

Search cost approaches zero for any activity that can be specified as a structured query over a known domain.

The implication for the firm: there is less reason to internalise capability simply because finding it externally was expensive. The “we hire because searching is hard” justification has largely evaporated.

2. Matching Costs — Compressed, Not Eliminated

Matching is subtler than search.

Search finds candidates. Matching determines fit — skill alignment, cultural compatibility, timing, information symmetry.

AI has compressed matching dramatically. Semantic similarity, embedding-based skill taxonomies, behavioural profiling, and structured interviewing protocols allow agents to perform first-pass matching at scale. What took weeks of human evaluation now takes minutes.

But matching has not been eliminated.

Deep matching — the kind that determines whether a partnership will survive ambiguity, whether a team will cohere under pressure, whether a contractor understands the unstated assumptions of a project — remains stubbornly human. It requires context that is not yet fully legible to automated systems.

Matching cost has split: surface matching approaches zero; deep matching remains expensive and irreducibly relational.

The firm’s role in matching has shifted. It no longer needs to warehouse talent to avoid repeated matching cycles. But it retains an advantage in maintaining the relational context that deep matching requires — shared history, embedded trust, accumulated working knowledge.

3. Contracting Costs — Automated at the Edges, Persistent at the Core

Contracts specify obligations.

For routine transactions, AI-generated contracts — templated, parameterised, validated against regulatory databases — have reduced contracting costs substantially. Smart contracts and protocol-layer agreements (Post 12) automate the mechanical aspects of specification.

But Williamson’s fundamental insight remains: contracts are incomplete.

The more complex the task, the more contingencies go unspecified. No contract can anticipate every state of the world. And when contracts are incomplete, someone must hold residual authority — the right to decide what happens in cases the contract does not cover.

This is precisely what protocols do within the networked firm. They standardise the foreseeable interactions and leave a governed space for the unforeseeable. But that governed space requires an institution — someone accountable, someone with standing to make binding decisions.

Contracting cost has fallen for routine transactions. But the cost of governing incomplete contracts — the residual — has not. It has merely moved from legal departments to protocol governance.

4. Monitoring Costs — Transformed by Observability

Monitoring was once the province of middle management.

Supervisors watched. Reports were filed. Quarterly reviews assessed performance against opaque criteria.

The shift is structural. In a protocol-governed, event-driven architecture (Posts 12 and 13), monitoring is not an additional activity. It is embedded in the system. Every state transition emits an event. Every agent action is logged. Every value stream step is observable.

Monitoring cost per transaction has fallen dramatically. But a new cost has emerged: the cost of interpreting the signal. Observability generates data. Making sense of that data — distinguishing meaningful deviation from noise, identifying systemic drift, recognising emergent failure modes — requires judgement.

Monitoring cost has shifted from observation to interpretation. The bottleneck is no longer seeing what happened, but understanding what it means.

This is where the Cyborg Cell (Post 9) earns its place. The human anchor in the cell is not monitoring in the traditional sense. They are interpreting the output of automated monitoring — exercising the judgement that closes the loop between observation and action.

5. Enforcement Costs — The Stubborn Residual

Enforcement is where the optimistic narrative of cost collapse meets institutional reality.

When things go wrong — when a supplier fails to deliver, when an agent produces a harmful output, when a protocol participant violates terms — someone must bear consequence.

AI has not reduced enforcement costs. In some cases, it has increased them.

Consider: when an autonomous agent makes a procurement decision that turns out to be fraudulent, who is liable? The agent has no legal standing. The protocol has no legal personality. The platform may disclaim responsibility.

Enforcement requires:

• Legal personality — an entity that can be sued
• Liability — someone who bears financial consequence
• Jurisdiction — a legal system that can adjudicate
• Accountability — a person who can be held responsible

None of these can be automated away. They are institutional primitives.

Enforcement cost has not fallen. It has been revealed as the irreducible core of institutional design — the cost that persists after all others have been compressed.

This is the strongest argument for the persistence of the firm. Not efficiency. Not coordination. Accountability. Someone must be the entity that answers when the system fails.

6. Managerial Bandwidth — Radically Expanded

The final cost — and in many ways the most transformative.

Post 1 noted that managerial bandwidth is finite. Each coordination edge consumes attention. Span of control is limited. As firms grow, communication paths multiply, monitoring load increases, decision latency rises.

AI agents have fundamentally altered this constraint.

A human manager with seven direct reports faces Brooks’s Law with every addition. An AI-augmented manager — operating within a Cyborg Cell, supported by agents that handle scheduling, summarisation, monitoring, and first-pass decision-making — can effectively coordinate across a much larger span.

But the expansion is not uniform. It follows the geometry of the network (Post 11).

In hierarchical networks, managerial bandwidth scales linearly. In connection networks, it scales with N². In group-forming networks — the topology that AI enables — it scales exponentially (Reed’s Law), but only if coordination is governed by protocols rather than direct supervision.

Managerial bandwidth has expanded — not by making managers work harder, but by changing the topology through which they coordinate. The protocol layer is the new span of control.

The Boundary Has Shifted — Unevenly

Here is the revised ledger:

The pattern is not uniform collapse. It is differential compression.

Some costs have approached zero. Others have transformed in kind. One — enforcement — has remained stubbornly high and may be growing as autonomous systems create new forms of liability.

The make-or-buy boundary has not simply moved inward or outward. It has become fractal — different for each cost component, different for each type of activity, different for each network geometry.

What Is the Firm For?

If search costs are near-zero, matching is automated, monitoring is embedded, and managerial bandwidth is expandable — why does the firm persist?

The naive answer is: it shouldn’t. If transaction costs are falling, Coase’s logic implies firms should shrink toward zero. Pure market coordination should dominate.

This answer is wrong.

It is wrong because it treats all transaction costs as equivalent. It assumes that compressing search and monitoring is the same as compressing enforcement and accountability. It is not.

The firm persists because it provides something that agent networks, protocols, and platforms cannot provide on their own:

Trust anchoring.

Someone must be accountable. Someone must hold liability. Someone must make the binding decision when the protocol encounters a case it was not designed for. Someone must answer to regulators, to courts, to the public.

That someone is the firm.

The Firm as Trust Anchor

The traditional firm performed many functions simultaneously: it searched, matched, contracted, monitored, enforced, and managed. It was a bundled institution — a single entity that handled all transaction costs internally.

The emerging firm unbundles most of these functions. Search is delegated to agents. Matching is performed by protocols. Monitoring is embedded in event architectures. Managerial coordination flows through protocol-governed networks.

What remains is the function that cannot be delegated:

• Liability — the firm is the legal entity that bears consequence
• Accountability — the firm is the institution that humans hold responsible
• Intent — the firm is the vessel through which human purpose is expressed
• Governance — the firm sets the rules that protocols encode

The firm’s new role is not to perform coordination. It is to anchor the trust that makes coordination possible.

This is a profound shift. The firm moves from being a coordination machine to being a trust institution. It still coordinates — but the coordination is increasingly performed by systems it governs rather than systems it operates.

The Firm as Protocol Governor

If the firm is a trust anchor, its operational expression is protocol governance.

Protocols (Post 12) define who can participate, what actions are valid, how outcomes are verified, and what happens when rules are violated. They are the coordination grammar of the networked firm.

But protocols do not write themselves. They do not adjudicate edge cases. They do not update when the world changes. They do not take responsibility when they fail.

The firm does.

The protocol governor:

• Designs the rules of interaction
• Admits participants to the network
• Audits protocol compliance
• Adapts protocols when conditions change
• Bears liability when the protocol produces harm

This is not a diminished role. It is a different one. The CEO of a protocol-governed firm does not manage people in the traditional sense. They govern a system — setting boundaries, defining acceptable behaviour, ensuring that the network produces outcomes aligned with human intent.

The firm does not disappear. It ascends — from operator to governor.

The New Boundary Condition

Post 1 stated the boundary condition as:

TC_market = TC_internal

This remains true. But the terms have changed.

TC_market is no longer dominated by search and matching. Those costs have been compressed by agents and protocols. What remains in TC_market is the cost of trust — the risk of transacting with parties whose accountability is uncertain, whose liability is unclear, whose intent is opaque.

TC_internal is no longer dominated by managerial bandwidth. That constraint has been relaxed by protocol governance and agent augmentation. What remains in TC_internal is the cost of maintaining institutional coherence — the overhead of being the entity that answers for the network.

The new boundary condition:

TC_trust_deficit (market) = TC_institutional_coherence (firm)

The firm expands to encompass activities where trust cannot be established through protocol alone — where human judgement, legal accountability, and reputational stakes require institutional backing.

The firm contracts where protocols and agent networks can establish sufficient trust through transparency, observability, and automated enforcement.

What Has Changed, What Has Not

Some things have not changed.

Coordination still has costs. Markets still have frictions. Institutions still exist to reduce those frictions. The logic of Coase — that the boundary of the firm is determined by the relative cost of internal and external coordination — holds.

What has changed is the composition of those costs.

In 1937, search and monitoring dominated. Firms existed primarily to avoid the expense of repeatedly finding and watching external parties.

In 2026, enforcement and accountability dominate. Firms exist primarily to provide the trust infrastructure that autonomous networks cannot generate on their own.

The firm has not been disrupted. It has been distilled.

Arriving

This series began with a simple observation: firms are coordination machines.

We traced the arc — from Coase’s insight through the managerial corporation, the limits of scale, the innovator’s dilemma, the digital boundary shift, the platform as proto-firm, agentic labour, knowledge diffusion, the hybrid topology, coordination density, network geometry, the protocol layer, value stream mapping, and the emerging agentic ecosystem.

Each post peeled back a layer of the question: what happens to the firm when AI changes the cost of coordination?

The answer is not that firms disappear. That was always the wrong prediction — the category error of confusing efficiency with institution.

The answer is that firms change function.

They move from being the locus of production to the locus of accountability. From managing work to governing protocols. From internalising capability to anchoring trust.

The corporation of 2026 is smaller in headcount but larger in reach. It coordinates more activity with fewer employees. It operates through networks rather than hierarchies. It delegates execution to agents and protocols while retaining the irreducible functions of governance, liability, and human intent.

The firm still exists because coordination has costs. But the nature of those costs has changed — and with it, the shape of the institution.

Coase was right. The question was never whether firms would persist. The question was what they would become.

Now we know.

References & Intellectual Lineage

• Coase, R. (1937). The Nature of the Firm.
• Williamson, O. (1985). The Economic Institutions of Capitalism.
• North, D. (1990). Institutions, Institutional Change and Economic Performance.
• Alchian, A., & Demsetz, H. (1972). Production, Information Costs, and Economic Organization.
• Jensen, M., & Meckling, W. (1976). Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure.
• Zuboff, S. (2019). The Age of Surveillance Capitalism — on accountability deficits in platform economies.
• Ostrom, E. (1990). Governing the Commons — on protocol governance as an alternative to hierarchy and market.
• Reed, D. (1999). “That Sneaky Exponential” — group-forming network geometry.
• Skelton, M. & Pais, M. (2019). Team Topologies — stream-aligned organisational design.
• Kim, J. & Liu, M. (2026). “Towards a Science of Scaling Agent Systems.” Google Research.
• Post 1 in this series: Why Corporations Exist — A Coasean Foundation.
• Post 9 in this series: The Hybrid Topology — the Cyborg Cell.
• Post 12 in this series: The Protocol Layer.
• Post 13 in this series: The Physics of Flow.
• Post 14 in this series: Value Stream Mapping.