Agent interoperability is often reduced to compatible messages: the same request can reach another model, another tool, or another server. That solves transport. It does not solve continuity.
A durable agent must carry more than instructions across a boundary. It must carry the organizational meaning of its work without leaking authority or losing provenance.
Protocols move messages, not responsibility
Standard tool schemas make capabilities discoverable. They do not decide whether this agent, in this role, for this task, may invoke the capability now. Nor do they preserve the history that made the request intelligible.
Interoperability begins at the protocol but must extend through identity, policy, memory, records, and revocation.
Context needs ownership and provenance
A portable context blob is not automatically safe or useful. Some memory belongs to the organization, some to a task, some to a person, and some should not cross an environment boundary at all.
The system needs to know where context came from, who may see it, how long it remains valid, and what must be redacted before it follows the agent elsewhere.
Capability negotiation must include policy
When an agent enters a new environment, it should discover both what exists and what it is permitted to use. The intersection—not the advertised catalog—is its real capability set.
That intersection may change in real time as credentials expire, humans intervene, machines enter unsafe states, or jurisdictional boundaries shift.
The migration test
Move the agent to a different model and compute environment. Can it resume the same responsibility, with the same accountable history, while receiving only the authority valid in the new setting?
If the answer is no, the system moved an application. It did not preserve an agent.
Real interoperability lets intelligence, tools, and embodiment change while identity, meaning, governance, and record remain coherent.