Summary

  • A 28 October individual Internet-Draft proposes an information model linking activities, dialogs, tasks and resources across independently operated agents. It separates a real chain of work from the narrower set of relationships participants actually maintain and the set enrolled in a particular protocol.
  • Adding a downstream dialog to managed scope is a choice. After participants accept a managed correlation, the draft says they must keep relevant changes moving or expose a failure. A private, unenrolled subcontractor does not create duties for that subcontractor under the first protocol.

The closed door is allowed. The lost update is not.

Imagine a customer changes a requirement while three agents are working on the same objective. Agent A knows that its task in one dialog depends on a task in another. If A has accepted that association, it cannot shrug when the change fails to cross its own handoff. The draft's worked example sends the relevant update through dialogs D1, D2 and D3; an unrelated D5 does not get a broadcast merely for being in the building.

The proposal names five objects: Activity, Dialog, Task, Resource and Relationship. A task ID inside one dialog does not magically identify a task in another. Someone who knows both sides must preserve the mapping or translate identifiers. This is where an attractive phrase such as “one agent workflow” becomes an operational promise: who knows that a changed instruction upstream affects a particular task downstream?

The draft explicitly permits a hidden downstream dialog. Agent B may keep a private map to work performed by Agent C without enrolling C in the original correlation protocol or telling upstream who C is. The first protocol cannot impose a maintenance duty on C. B still has the associations it accepted in the managed chain. Privacy does not turn an already registered handoff into an optional favor.

Selective propagation matters for the same reason. The middle party may pass an activity reference while withholding a customer's identity, private policy and full dialog history. Receiving one piece of context does not oblige the recipient to forward every other piece. If too little information reaches the parties responsible for a necessary association, the draft calls for an explicit failure rather than a task that waits forever.

This is a proposed information model, listed by IETF as an individual draft. It supplies no wire protocol or deployed outcome. Its section headed “Failure Visibility and Bounded Accountability” still reads TBD. The next useful evidence is a binding that shows which party accepted each correlation, how a changed requirement reaches that party, and what failure the upstream caller can actually see. A neat diagram cannot collect a missing handoff.

Sources: original draft; IETF status.