Why more than one agent
A single agent juggling many roles bloats its context and blurs its focus. Multi agent systems split work across specialized agents, each with its own tools and prompt, coordinated toward a shared goal.
Common topologies
- Orchestrator workers a lead agent decomposes the task and delegates to workers
- Pipeline agents run in sequence, each refining the previous output
- Debate agents argue toward a better answer, a judge decides
- Blackboard agents read and write a shared workspace
Orchestrator pattern
A lead agent plans, spawns workers for subtasks, collects their results, and synthesizes a final answer. Workers stay focused on narrow jobs.
The hard parts
Coordination adds overhead. Agents can duplicate work, contradict each other, or pass along errors. Clear role boundaries, a single source of truth for shared state, and a synthesis step that reconciles outputs keep the system coherent. More agents is not always better.
Key idea
Multi agent systems trade single agent simplicity for specialization, but only pay off when roles are clear and a synthesis step reconciles their outputs.