multi-agent system
Several agents with separate contexts and roles working on one task, coordinated by a lead or by handing work to each other.
In more detail
Anthropic's research system found multi-agent runs used around fifteen times the tokens of a chat. Start with one agent and split only when the work divides cleanly and the gain pays for the tokens.
Related terms
- sub-agent — An agent started by another agent to do one piece of work in its own fresh context, returning only its finding.
- orchestrator-workers — A pattern in which a lead model breaks a task into subtasks at run time, hands them to worker calls or sub-agents, and combines their results.
- AI agent — A system in which a model decides its own next steps, calling tools in a loop until a task is done, rather than following a path your code fixed.
Taught in
- When Context Fails — AI Engineering: LLM Apps, RAG & Agents
- Workflows Before Agents — AI Engineering: LLM Apps, RAG & Agents
- Planning, Memory And Many Agents — AI Engineering: LLM Apps, RAG & Agents
References
- Anthropic: How we built our multi-agent research system
- OpenAI Agents SDK: Running agents (the agent loop, max_turns)