Agentic commerce

Agentic commerce needs settlement infrastructure.

Autonomous agents will increasingly initiate payments, negotiate terms, and settle obligations. That makes the bilateral question more urgent, not less: when two agents transact, there must be a signed record both principals hold — and any third party can verify.

Attribution is the precondition

CAST already requires every event to declare its actor mode — human, agent, system, or counterparty — and to name the accountable human owner whenever an action can change financial state. This is exactly the control structure agentic commerce needs: an agent can execute, but a human remains accountable, and the record proves which was which.

An agent can be the performer. A human must remain the accountable owner. The decision record keeps the two distinct — permanently, and verifiably.

The agent loop, gated

Agents propose; the gate stays bilateral

An agent can draft and route a payment, but a covered event still requires counterparty co-authorship before value moves.

Policy as machine-checkable specification

Proposed policy changes can be expressed as formal state-machine specs, model-checked against all reachable states, and shadow-executed against history before any human approves them.

Proof travels with the transaction

Whatever produced the event — human or agent — the lineage hash and bilateral confirmation make it independently verifiable downstream.

Why this strengthens the case

The agentic future does not weaken the argument for bilateral proof. It is the argument for bilateral proof. As execution gets cheaper and faster, the only durable asset is the ability to prove, years later, that an event occurred and what it contained. The more autonomous the actors, the more that proof is worth.

Humans buying from agents, agents buying from agents

How accountability is preserved when neither party is a person.