A Thousand Contracts, One Principal

The nervous system and the automatic machine are fundamentally alike in that they are devices which make decisions on the basis of decisions they made in the past.

Coase asked why some transactions are coordinated through markets and others inside firms.¹ His answer began from the cost of using either arrangement. Search, negotiation, monitoring, and enforcement make exchange expensive. Administration is expensive too. A firm exists in the interval between those costs.

Computational systems alter that interval unevenly. They can search a thousand offers, draft a thousand instruments, monitor a thousand deliveries, and initiate a thousand transfers. The number of principals need not increase at all.

A Situated Actor

The word agent invites an argument about personhood before the economic work has been described. A practical actor requires much less. Give a computational process an objective, access to records, tools it can invoke, and discretion to select among actions in a particular case, and part of judgment has been delegated unfinished. The process need not desire the outcome or own the resources it directs.

Such an actor may be brief. A runtime is invoked with a model, instructions, context, credentials, and tool bindings. It selects and acts, then terminates. Memory and identity can persist through stores, keys, logs, and accounts, but those components belong to an architecture maintained by an institution. The situated actor and the surviving principal are not the same thing.

This distinction changes the Coasean problem. A firm can reduce the marginal cost of search and drafting by deploying many computational actors while preserving one treasury, one policy, and one legal identity. It can also expose itself to a new kind of scale: more commitments can be initiated than any human office could inspect one by one.

Market Entry

Consider a constructed procurement case. An actor working for a manufacturer searches supplier records, compares prices and delivery windows, drafts a purchase order, and routes payment under a delegated limit. The cognitive sequence may finish in seconds. The trade still depends upon an identity the supplier recognizes, authority to bind the buyer, a settlement rail, evidence of delivery, and recourse if the goods fail specification.

Those requirements attach at different places. A certificate may authenticate a key without proving that its holder was authorized for this purchase. A bank or protocol may move funds without establishing that the shipment conformed. A sensor may record arrival without interpreting a warranty. A bond may add assets to a remedy without deciding who was at fault. Market entry is a composition of interfaces, not a single credential.

The computational actor can cross some of them directly and others only through a principal. If the order falls within a predeclared policy, the institution may treat the actor's signature as its own authorization. If a dispute follows, the vanished runtime is not the only object available. Policies, logs, model versions, tool calls, accounts, and officers can preserve an evidentiary path to the organization that designed the delegation.

A thousand contracts can therefore have one principal. That economy is the source of both scale and danger. Responsibility has not escaped humanity. It has been concentrated upstream, in the decision to define the actor's objective, scope, evidence, and power.

What Changes in the Firm

Williamson identified bounded rationality, opportunism, and asset specificity as pressures shaping governance.² Computation does not erase any of them. It redistributes their costs.

Search across more alternatives can reduce one form of bounded rationality while creating dependence on training data, interfaces, and specifications. A computational actor has no private appetite for gain, but a principal can use it opportunistically, and an objective can exploit what its designers failed to specify. Asset specificity remains wherever a plant, dataset, workflow, or relationship has more value inside one arrangement than outside it.

The familiar prediction that cheaper contracting makes firms disintegrate is consequently incomplete. Standardized, easily checked services may move toward markets. Activities requiring proprietary context, licenses, physical integration, liability capacity, or continuing judgment may move inward. The same technology can thin one boundary and thicken another.

V/C can help locate the pressure, provided it remains a heuristic. Cheap verification relative to the value at stake can support more delegation and finer-grained exchange. High expected loss, difficult remedy, weak evidence, regulation, or scarce authority can dominate that comparison. Verification cost is one term in the decision, not the law that draws the firm.

Authority, Liability, Settlement

Authorization and liability are often joined, but they are not identical. A professional license can authorize an act while an employer or insurer bears much of the loss. A corporate policy can authorize a purchase while different officers, entities, and vendors divide responsibility. The institutional problem is to prevent those divisions from dissolving the claimant's route to remedy.

A cryptographic signature proves control of a key under a scheme. It does not by itself prove identity, authority, purpose, or legal recognition. Conversely, a human signature is not magically answerable. The important object is the chain joining an act to a principal that can explain the delegation and to a forum capable of changing the consequence.

Settlement is plural for the same reason. Bank money, platform credit, stablecoins, Bitcoin, and secured legal claims offer different combinations of stability, access, issuer exposure, custody, auditability, finality, and recourse. No asset follows logically from the fact that software initiated a payment. Chapter 15 identified the unusual office performed by proof-of-work. It did not make that office necessary for every transaction.

At software speed, computational actors may form dense markets for search, information, and bounded digital service. At institutional speed, principals still obtain licenses, insure risks, build plants, submit evidence, and answer claims. The interface between those tempos is not destined to belong to one universal platform or protocol. It is a field of economic competition and constitutional choice.

Coase's question thus survives with a new term. The cost of using the market now depends not only on finding and bargaining with another party, but on making a computational act answerable across identity, authority, evidence, settlement, and remedy. Where that composition becomes cheap, exchange can multiply. Where it remains scarce, the institution that supplies it may become the firm, the gatekeeper, or both.