Agent Swarm
A group of AI agents that work together — often with different roles — to solve a problem one agent could not handle alone.
In plain English
An agent swarm is a multi-agent system where several specialised agents collaborate, debate, or vote on a task. Instead of asking one large model to do everything, you spin up smaller agents with focused roles — a researcher, a writer, a critic — and let them coordinate.
Why use a swarm:
- Specialisation — each agent gets a tighter prompt and toolset
- Self-correction — a critic agent can catch errors before output
- Parallelism — independent sub-tasks run at the same time
- Resilience — one agent's failure doesn't kill the whole run
Examples: OpenAI's Swarm framework, CrewAI's "crews," AutoGen's group chat, and the Swarms.ai library all implement this pattern. Production examples include research agents that split queries across sub-agents, and coding agents that run a separate review pass before committing.
Swarms add complexity and cost — most teams start with a single agent and only graduate to a swarm when one agent can't reliably hold the whole task.