Multi-Agent System
In one line
A multi-agent system is several AI agents with different roles cooperating on the same task — used when one agent alone is not enough to solve the problem.
Going deeper
A multi-agent system is several agents with different roles — Researcher, Writer, Critic — collaborating on one task. It usually beats a single agent on quality because each step is specialised, and the agents catch each other's mistakes.
A common marketing automation lineup is researcher, copywriter and brand-guide reviewer. Token cost and latency go up versus a single agent, but the variance in output quality drops a lot, which is often what teams actually need.
The downsides are real. More agents make debugging harder, and they sometimes get stuck pinging each other forever. The standard fix is a dedicated Orchestrator that steers the flow and enforces explicit stop conditions.
Related terms
Orchestrator Agent
An orchestrator agent is the top-level agent that schedules sub-agents, assigns roles and enforces stop conditions — the control tower of a multi-agent system.
AI AgentA2A
A2A (Agent-to-Agent Protocol) is the agent-to-agent standard Google launched and donated to the Linux Foundation in mid-2025, letting agents from different vendors delegate work to each other and exchange results.
AI AgentPlanning Agent
A planning agent is an agent that first drafts a multi-step plan for a goal and then executes it step by step, calling tools as needed.
AI AgentAI Agent
An AI agent is an LLM-driven system that takes a goal, plans the steps, calls the tools it needs and runs the task end-to-end with limited human input.
AI AgentAgent Evaluation
Agent evaluation is the test and metric framework for measuring how accurately and safely an agent completes its goals — distinct from plain LLM benchmarking.