handoff
In plain wordsThe moment one AI agent passes a conversation or task to another, better-suited agent — like a call-center operator transferring you to the right department.
In multi-agent systemsmulti-agent system, n.: A setup in which several AI agents, each with its own role or point of view, work together — or compete — on a problem, passing messages and results among themselves., the transfer of control over an ongoing conversation or task from one agentAI agent, n.: An AI system that does not just answer but acts: given a goal, it decides on steps, uses tools such as a browser or code editor, checks the results and keeps going until the job is done. to another, together with the conversation history or a summary of it. The receiving agent continues with its own instructions and tools, and the first agent does not resume unless control is passed back. Handoffs are usually implemented as a special tool calltool call, n.: A single request from an AI model to use one of its tools — naming the tool and filling in the details — which the surrounding software then carries out and reports back on.: the active agent calls a transfer function whose return value is the next agent. They suit triage designs — a front-line agent routing billing questions to a billing agent and technical ones to a support agent — and escalation to human staff. Design questions include how much context to pass (full history or a summary), how to prevent agents bouncing a user back and forth, and how to keep the transition seamless for the user.
The triage agent hands off to the refunds agent once it has the order number.
Origin
Borrowed from telephony, relay races and clinical shift changes. OpenAI's cookbook 'Orchestrating Agents: Routines and Handoffs' (Ilan Bigio, 10 October 2024) and the accompanying experimental Swarm library made 'handoff' a named primitive; it was carried into the OpenAI Agents SDK (March 2025) and adopted by other agent frameworksagent framework, n.: A software library that gives developers ready-made parts for building AI agents — model connections, tool handling, memory, multi-agent coordination and logging — so they need not write everything from scratch..
Usage note
Contrast with calling a subagentsubagent, n.: A helper AI agent that a main agent starts to handle one piece of a job; it works in its own separate workspace and reports back a short summary when it is done., where the caller keeps control and receives a result. In customer-service automation 'handoff' also means passing a case from an AI agent to a human.
Related
- multi-agent systemmulti-agent system, n.: A setup in which several AI agents, each with its own role or point of view, work together — or compete — on a problem, passing messages and results among themselves.
- subagentsubagent, n.: A helper AI agent that a main agent starts to handle one piece of a job; it works in its own separate workspace and reports back a short summary when it is done.
- orchestrationorchestration, n.: Coordinating several AI calls, tools or agents so each does its part in the right order and their results are combined into one finished piece of work, as a conductor directs an orchestra.
- agent frameworkagent framework, n.: A software library that gives developers ready-made parts for building AI agents — model connections, tool handling, memory, multi-agent coordination and logging — so they need not write everything from scratch.
- tool calltool call, n.: A single request from an AI model to use one of its tools — naming the tool and filling in the details — which the surrounding software then carries out and reports back on.
- subagentsubagent, n.: A helper AI agent that a main agent starts to handle one piece of a job; it works in its own separate workspace and reports back a short summary when it is done.
Sources
- Bigio, I. (2024). Orchestrating Agents: Routines and Handoffs. OpenAI Cookbook, 10 October 2024. developers.openai.com/cookbook/examples/orchestrating_agents
- OpenAI (2025). New tools for building agents. 11 March 2025. openai.com/index/new-tools-for-building-agents
Mentioned in
agent frameworkagent framework, n.: A software library that gives developers ready-made parts for building AI agents — model connections, tool handling, memory, multi-agent coordination and logging — so they need not write everything from scratch., Agent2Agent protocolAgent2Agent protocol, prop. n.: An open standard that lets AI agents built by different companies find each other, say what they can do and work together on tasks, much as email lets different mail systems exchange messages., multi-agent systemmulti-agent system, n.: A setup in which several AI agents, each with its own role or point of view, work together — or compete — on a problem, passing messages and results among themselves., orchestrationorchestration, n.: Coordinating several AI calls, tools or agents so each does its part in the right order and their results are combined into one finished piece of work, as a conductor directs an orchestra., subagentsubagent, n.: A helper AI agent that a main agent starts to handle one piece of a job; it works in its own separate workspace and reports back a short summary when it is done., agent swarmagent swarm, n.: A large group of AI agents working at the same time on parts of a problem, usually with little central control — a name loosely borrowed from swarms of insects.