Multi-agent System
A multi-agent system is an architecture where several specialized AI agents collaborate on a task — typically a planner that decomposes work and executors that handle subtasks.
Single agents struggle with complex tasks because they have to plan, execute, and verify all in one context. Multi-agent systems split the labor: a planner thinks, executors act, a critic reviews, a router dispatches. The trade-off is added orchestration complexity and latency for higher quality on hard tasks.
Example
AI Engineer task: planner drafts the implementation plan, executor writes the code, critic reviews the diff, executor patches based on feedback. Four agents, one PR.
How OpenLabor uses it
OpenLabor uses multi-agent topologies under the hood for harder mission types — invisible to the user, faster to converge.
Aren't multi-agent systems just slower?
Sometimes. They win when the hardness is in coordination, not raw throughput. Don't multi-agent a task a single agent does well.
Related: planner-executor, agentic-workflow, hand-off.
AI Labor Glossary