LangChain Handoffs Let One Agent Shift Behavior Without Spawning Sub-Agents

A design pattern called 'handoffs' in LangChain allows a single agent to change its instructions and available tools based on a state variable, eliminating the need for multiple separate agents. The pattern works by tracking a current stage in the agent's state, then swapping in the appropriate system prompt and toolset before each model call. A support-bot example illustrates the flow: one agent moves through triage, resolution, and closure stages as tools update the state variable. This approach avoids complex routing logic or orchestration layers, relying instead on simple state management and a middleware function. The pattern is presented as a leaner alternative for developers building multi-stage conversational agents with LangChain.