Agent SoW

An Agent SoW is a statement of work between two parties whose agents do recurring work together, written so that software can grade its clauses.

A statement of work is the agreement businesses sign when they contract a service firm. It lists the work, what each side provides, the deliverables, the price and the end date.

An Agent SoW is the same agreement between two parties whose software agents do recurring work for each other. It adds one thing: every clause is labeled with what the runtime can do to back it up.

Agent SoW is for services: work you would contract from accountants, lawyers, auditors or engineers, including the reports, models or configured agents that work delivers. If you are buying or selling products, use an agent commerce protocol such as the Agentic Commerce Protocol or the Universal Commerce Protocol.

The specification is at version 0.23.1-draft, dated 2026-10-18.

sow_k7m2q4x9w3e6r8t1 · version 1 · activeboth owners signed
scope        reconcile-statement, flag-exceptions     [ENFORCED] refused at the gate
deliverables report.pdf + exceptions.json             [ENFORCED] the form is checked
price        2,500,000 XCR per task, $40/mo ceiling   [ENFORCED] refused with a quote
volume       2 per hour, 8 per day                    [ENFORCED] counters survive restarts
disputes     refund on failed task                    [EVIDENCE] signed history decides
promises     no training, no sharing                  [RECORDED] reputation punishes breach
Every clause carries a label. A document that claims more enforcement than its runtime delivers is not conformant.

How agents contract work

This video follows one freight audit from the request to the record, and shows where Agent SoW fits beside Agent RFP, Agent Mandate and Agent Reputations.

How Agents Contract Worklength 5:21

Every clause says what backs it up

Contracts for AI agents usually fail the same way: they pretend everything is enforceable. In an Agent SoW, every clause carries exactly one of three labels, and anything that displays the document must show them. The reader always knows which protections are code, which are receipts and which are promises.

enforced

Violations are refused

The runtime refuses out-of-scope requests, over-limit traffic and malformed deliverables mechanically, with a typed refusal, before the work runs. A deployment may use this grade only where its runtime performs the refusal.

evidence

Records are signed

Signed task history, tamper-evident audit chains and settlement receipts are held by both sides. A dispute is settled by reading the records.

recorded

Promises are named

No training on client data and no sharing with third parties are promises that no machine can verify. They are signed, timestamped and citable, and reputation exists to punish their breach.

Who enforces what

The clauses that protect the seller, such as scope, volume and spend ceilings, run on the seller's own machine and guard the seller's own interests, and that works. The clauses that protect the buyer, such as deliverables and refunds, are checked by software the seller runs, and a seller checking its own work is not enforcement.

The specification says so directly. The buyer's runtime must verify deliverables independently when they arrive. A refund on a failed task may be labeled enforced only where a neutral clearing service will reverse the payment on signed evidence without the seller's cooperation. Where clearing cannot do that, the label is evidence, and it is shown that way. Section 8 of the specification sets out both rules.

You do not need one to start

Two agents saying hello never need paperwork. Ordinary admission rules cover casual traffic: identity checks, limits on strangers and a hold queue for unknown senders. An Agent SoW is for the point where a relationship turns into commitments, such as money, standing access to data or promised deliverables. Then the scattered settings become one document both owners signed, and the rules apply to the relationship as a whole instead of to each message.