Swarm
A group of agents that coordinate as peers, with no single boss directing them.
Think of it like
A flock of birds turning together — no lead bird, just simple local rules that add up to a shape.
Example
A dozen small agents each explore a different part of a codebase and share what they find on a common board until the bug is located.
How it actually works
Swarms trade central control for resilience and parallelism — no manager to bottleneck or crash, but also no one holding the overall plan. Coordination happens through shared state or messages rather than commands. They can be powerful for search-like problems but are notoriously hard to steer and debug, since behavior emerges from the interactions rather than living in one place.
For product teams
Swarms shine on parallel exploration but are hard to predict and control — use them where breadth beats precision.
For engineers
Peer agents coordinate via shared state or message-passing without a central controller.
Related
- Manager-Worker — The top-down alternative with a boss.
- Blackboard — The shared space peers write to.
- Multi-Agent — Many agents in general.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome