Creativity · Agent Protocol

Handoff vs Delegation — A2A Semantic Distinction

In multi-agent systems, handoff and delegation are often used interchangeably — but they describe different control-flow semantics. A handoff transfers the conversation to another agent; the caller is done. A delegation assigns a sub-task to another agent but keeps the caller alive, waiting for a result to integrate. Picking the right one matters: handoffs are cheaper but lose context, delegations keep the caller's plan intact but cost more.

Protocol facts

Sponsor
open pattern (OpenAI, Anthropic)
Status
stable
Spec
https://www.anthropic.com/research/building-effective-agents
Interop with
OpenAI Agents SDK, Claude subagents, LangGraph, A2A

Frequently asked questions

Which should I default to?

Start with delegation. It's easier to reason about — your orchestrator keeps the plan and integrates results. Move to handoffs when the next specialist genuinely owns the rest of the task end-to-end.

Can handoffs come back?

In some frameworks, yes — a handoff can be 'returnable' where the callee hands control back to the caller when its phase completes. That's effectively a delegation wearing a handoff's clothes.

How does A2A model both?

A2A's task lifecycle is closer to delegation by default — the caller keeps a task handle and can listen for updates — but you can build handoff semantics on top by having the callee open new tasks rather than continuing the original.

Sources

  1. Anthropic — Building Effective Agents — accessed 2026-04-20

Related