An agent handoff transfers control of a conversation or task from one AI agent to another specialist, or from an agent to a person. A useful handoff carries the goal, relevant context, completed work, unresolved questions, and authority boundaries the receiver needs to continue.
The OpenAI Agents SDK represents handoffs as tools that transfer execution to a named agent. Customer service systems use the same pattern when automation escalates to a human operator.
When should one agent hand work to another?
Use a handoff when a specialist should own the next decisions, not merely answer one question. Ownership transfer is the defining difference from the orchestrator-worker pattern, where the orchestrator keeps control and assembles worker results. Multi-agent systems can use either model.
What makes a handoff reliable?
The receiver needs source locations and current state, not a vague summary. The handoff must also preserve permission limits, especially when a human approval gate is still pending. Remote agents may advertise their capabilities through an agent card and communicate through the A2A protocol.
Frequently asked questions
Is an agent handoff the same as delegation?
No. A handoff transfers control to the receiver. Delegation can keep the original agent in control while another agent performs one bounded task and returns a result.
What information should an agent handoff include?
It should include the goal, verified facts, completed actions, current state, open questions, relevant source locations, and any permissions or limits that remain in force.
No advertising or tracking cookies, and our visitor counts are anonymous. The Cal.com booking widget loads only if you allow it. Privacy Policy.
The page itself, anything our host sets to serve and secure it, and the anonymous visitor count. Always on, and none of it stores anything on your device.
The Cal.com booking widget. Left off, a booking link opens the booking page instead of a popup, so you can still book a call.