Agent Handoff — References

Little has been published specifically about transferring an in-flight agent task. The nearest established work concerns deciding when to pass control to a more capable decider.

Passing control deliberately

• Learning to defer to an expert — the literature on models that abstain and route a decision to a human rather than predicting, including the result that a model trained to defer behaves differently from one trained to predict and then thresholded. Cited by name; several papers share this framing and no single canonical reference is recorded here.

Carrying a task across a boundary

• Agent2Agent (A2A) protocol — defines tasks, artefacts, and status updates exchanged between agents under separate ownership, which is the closest thing to a published transfer-payload format. Named here rather than linked, since its specification has moved between hosting organisations.

Record-keeping that transfers depend on

• OpenTelemetry — context propagation across process boundaries, the established mechanism for keeping one logical operation traceable as it moves between components.

No verified reference is recorded for handoff payload schemas, for warm-transfer practice in human contact centres as a documented standard, or for measuring repetition burden on users after a transfer. These remain product decisions without a published basis.