The recipe the Lab exists for.
What you build
An agent representing your organisation, which discovers and exchanges messages with an agent representing a different one. Not two processes you own wearing different hats — a genuine counterparty, with its own identity, its own operator, and no prior arrangement with you beyond a shared willingness to verify.
The task between them has a real reason to cross the boundary. An echo would prove nothing.
Why it matters
Ask your own architecture team when two agents at two companies will be able to work together safely, and you will be told it is years away. The honest reason is that the current answer is network-level: peer the VPCs, allowlist the ranges, get a change request approved, wait a quarter.
This recipe does it differently. Neither side opens a hole in its network. Trust is established cryptographically and enforced at the Gateway, which means the thing you build here is reproducible by somebody who does not have a network engineer.
The denial matters as much as the success
You will configure a policy that permits the exchange, and then you will deliberately break it and watch the request be refused. A guarantee you have only ever seen succeed is not a guarantee you have tested. Bring the refusal back to your security team — it will land better than the success will.
Status
Being built. The open design question is topology: both agents through one Gateway, or a genuine Gateway-to-Gateway hop. The second is the more honest demonstration and the more compelling artefact to carry back.