Home › Agents & Tool Use › Key term › Multi-agent
Key term · Build

Multi-agent

Several specialized agents divide a task and coordinate toward one outcome.

In one line

Multi-agent systems split hard work across cooperating specialist agents.

DefinitionWhat it means

A multi-agent system decomposes a task across several specialized agents, for example a planner, a researcher, and a coder, that communicate through messages, shared memory, or a supervising orchestrator agent, rather than relying on one general-purpose agent to do everything.

Why it mattersWhy you should care

Splitting responsibilities improves reliability and lets each agent use a narrower, cheaper, or more specialized model, but it introduces new failure modes such as coordination overhead, conflicting sub-goals, and higher latency and cost, all of which product teams must weigh against a simpler single-agent design before adopting the pattern.

At a glanceSee it

Multi-agent diagram
Multi-agent diagram 1

Real multi-agent work is a verify-and-retry loop rather than a straight line — a checker bounces weak drafts back until they pass, and a retry cap escalates a stuck worker instead of looping forever.

Multi-agent diagram 2

Choosing a coordination topology is itself a decision — driven by whether subtasks are independent and whether answers need cross-checking before they ship.

Where you see itIn the wild

  • Frameworks like AutoGen and CrewAI built specifically for agent teams.
  • Production support and coding assistants that route sub-tasks to specialist agents.
  • Design debates on when multi-agent design beats a single strong agent.

What changedWhat changed here

RecentAuto-linked from the brief, not a rewrite of this page

Three kinds of claim, strongest first. Signal runs every morning.

A living map of modern AI — kept current every morning