Multi-agent
Several agents with separate roles working on one task, often with one coordinating the others.
A multi-agent system splits work across several agents with different roles: one plans, one writes code, one reviews, one runs the tests. Sometimes a supervisor agent routes work between them.
The appeal is parallelism and specialisation. A reviewer agent with a fresh context, for example, catches things the implementer cannot see because it is too close to its own work.
The cost is coordination overhead. Every hand-off between agents costs tokens, loses some shared state, and can introduce a mistake at the seam. Many teams find that a single agent with well-designed tools and a clear loop beats a swarm of agents that talk to each other. Add agents when the roles are genuinely different, not when the task merely feels big.
Related terms
Agent
A model with tools, a system prompt, and a loop that lets it act and observe results.
Subagent
An agent spawned by another agent to do one scoped job in its own context.
Agent Loop
The repeated cycle of decide, act, observe. Where autonomy actually comes from.
Context Engineering
Deciding what goes into the model's context, and what stays out.
Harness
Everything around the model that turns it into an agent: tools, system prompt, permissions, hooks.
Want this working inside your team?
It starts with one conversation about your workflow.