Exchange standards as the contract
The interoperability formats are the starting specification, not an export format bolted on at the end. Conformance is measured against them.
Practice
Healthcare agents
Healthcare agents stand or fall on standards and interoperability. An agent that cannot speak the exchange formats, carry a scope of consent, or leave an audit record is not deployable, however well it talks.
Four requirements we treat as testable
The same approach as the commerce work: turn each obligation into a statement, then into a check that is proven to catch its own violation.
Consent scope, recorded and honoured
Identity, what it authorises, and for how long. An action outside the recorded scope has to be refused, and the refusal has to be visible.
An audit trail that reads back
Every action the agent took, with what it saw and what it changed, in a form that answers a question months later.
Refusal as a feature
The safe answer is often no. An agent that cannot decline, escalate and say why is a liability, so we test the refusals as carefully as the successes.
Why this practice sits beside the commerce work
Agentic commerce and healthcare exchange have the same shape of problem. Both are multi party. Both carry an obligation that outlives the request. Both fail on the contract long before they fail on the model.
The method transfers directly: read the specification, implement until it breaks, turn each obligation into a check, and ship the suite so the claim can be verified by whoever has to sign it off.
How an engagement starts
- A conformance assessment of the integration you already have, against the exchange standards it claims to meet
- Consent scope and audit obligations written as numbered statements
- A runnable suite, so the assessment can be repeated by your own team
- Then, if it is wanted, the build