Multiple specialised agents working on a task in parallel — orchestrated, not one generalist.
Category · AI & Agents
Many agents, one goal.
An agent swarm is a team of specialised agents working on a task together. An orchestrator distributes sub-tasks, sub-agents return results in parallel, a review step consolidates. The principle scales horizontally: more specialists instead of one generalist.
When that makes sense.
When a task clearly decomposes into sub-steps and those sub-steps are heterogeneous — a researcher, a calculator, an author, a reviewer. For simple, linear tasks a single agent is faster and cheaper.
Frameworks we use: LangGraph, CrewAI, Mastra, AutoGen and the Anthropic agents SDK. The choice depends on the product, not the brand.
