The Exchange: When Two AI Agents Meet | Geometry of Trust | Protocol - Lesson 3

Attestations and chains exist to be exchanged. The exchange protocol is the handshake in which every other layer of the stack finally comes together — measurement from the mathematics series, structural boundaries from the governance series, attestations and chains from the earlier protocol talks. All of it converges in a five-step exchange that either succeeds cleanly or fails with an auditable reason.

This talk walks through that handshake with a concrete running example: Alice (a diagnostic AI) meeting Bob (a drug-interaction checker). Both sit in healthcare but in different sub-domains, with asymmetric interaction modes (advisory ↔ read-only).

Step 1 is the domain check — exclusions, permissions, mode compatibility — run before any cryptography. Steps 2 and 4 are the messages on the wire: agent ID, signed envelope, attestation chain, current attestation, and (for the response) verdict. Steps 3 and 5 are the nine cryptographic-and-governance checks each side runs against the other, in a specific order designed to fail cheaply when it can.

The envelope is the one piece worth pausing on. An attestation by itself says "these were my readings at this moment" — a statement about the past. Without the envelope, an attacker could replay a genuine old attestation in a new conversation. The envelope binds the attestation to this specific exchange: a unique nonce, the current timestamp, the peer's identifier. Non-replayable.

Neither agent trusts the other until both have produced evidence and both have verified it. That's the symmetry. It's what makes this genuinely trustless cooperation — no central arbiter, no reliance on reputation, no trust-by-default. Two AI agents who don't know each other can still reach a verified, governance-enforced agreement to cooperate, or an auditable refusal not to.