The Agentic Boundary of the Firm

The boundary has a cost

A company coordinates some work internally and obtains other work through suppliers, partners, and markets. The boundary between those arrangements reflects more than price. It also reflects the cost of finding a capable party, specifying the work, exchanging information, monitoring performance, resolving disputes, and bearing failure.

Software has already reduced parts of that cost. Digital procurement, application interfaces, and shared records allow firms to exchange requests and track transactions. Agents may extend that capability into less structured coordination, where each request requires interpretation and several parties must maintain a shared understanding of what was promised.

If delegated software lowers the full cost of coordinating across companies, some work may become easier to organize externally. That is a conditional economic proposition, not a prediction that firms will simply become smaller.

A commitment across companies

Consider a manufacturer whose supplier reports that a component will arrive late. The manufacturer needs to understand which orders are affected, whether another supplier has capacity, what the contracts permit, and which customers have commitments at risk.

Today, people may assemble this picture through calls, messages, and separate systems. Authorized agents could exchange structured requests and evidence: the affected quantity, available alternatives, delivery windows, contractual constraints, and proposed changes. Each firm could keep its own records and decision authority while coordinating around a shared commitment.

The interaction would need more than agents that can converse. The parties must know which organization an agent represents, what it may disclose, what it may offer, and when an exchange becomes binding. They need records that both sides can inspect and a process for handling conflicting accounts.

An agent could accelerate the discovery of options. Accepting a revised price, disclosing sensitive capacity, or changing a customer promise would require the authority assigned to that decision.

Trust is part of the architecture

Inside a company, agents already face inconsistent data and divided authority. Across companies, those conditions intensify. Participants have different incentives. Their records may use different meanings for “available,” “delivered,” or “accepted.” Neither side can assume that the other’s agent is reliable or empowered to make a commitment.

Controlled interaction therefore requires identity, scoped delegation, permissions, provenance, audit, and mechanisms for revocation and dispute. Existing interfaces and identity systems provide pieces of this foundation. They do not, by themselves, settle commercial meaning, liability, or enforceability.

A practical design might begin with narrow exchanges: request a status, provide specified evidence, or propose a change for human approval. Broader autonomy should depend on demonstrated reliability and explicit contractual arrangements.

The economic question remains open

Lower transaction costs can support more external coordination, but they can also help large firms coordinate more activities internally. A company with superior data, infrastructure, and bargaining power may gain scale advantages from agents. Another may use them to work effectively with specialized partners. The outcome depends on which costs fall, for whom, and by how much.

Verification can erase apparent savings. A fast negotiation is of little value if each party must spend more time checking claims or repairing errors. The relevant measure is the cost of an accepted outcome, including integration, oversight, failure, and dispute resolution.

Firm boundaries are also shaped by capabilities and control. A company may retain work internally because it needs to protect knowledge, assure quality, or govern a critical dependency, even when an external transaction becomes easier.

A conditional network

Current agents can assist with bounded tasks inside and between organizations. An emerging architecture could let them exchange evidence and proposals under explicit authority. A plausible next state is a network of firms that maintain selected commitments through controlled software interaction.

A more fluid economic structure is a longer term possibility. It would require dependable cross company semantics, enforceable commitments, trusted evidence, and workable liability arrangements. These are institutional achievements as much as technical ones.

The first question for an executive is concrete: which costly interaction with another organization could become clearer, faster, and more reliable if both parties could maintain an explicit commitment through software?

The executive question

Which external relationship carries the greatest coordination cost today, and what evidence, authority, and commercial agreement would be required before agents could manage part of it?