AI glossary
Multi-agent system
A setup where several AI agents, often with different roles or tools, work together on a task.
A common pattern is an orchestrator agent that splits a big job into parts and hands them to worker agents — for example, several researchers each reading different sources, then a writer combining their findings. This allows parallel work and keeps each agent’s context focused.
More agents also mean more cost and more ways to fail, so start with a single agent and add more only when it clearly helps.
Example: Anthropic ran about 950 Claude agents in parallel for 21 hours to search a huge DNA database, and they found an enzyme system no one had described.
In practice
- It works best when the work splits into independent parts, such as researching many sources.
- It works worse when every agent needs the same context, as in many coding tasks.
- Anthropic says a multi-agent system uses about 15 times more tokens than a chat: save it for tasks that are worth it.
More in agent design patterns and in the enzyme story.
