Fan-out / fan-in pattern
Parallel Multi-Agent System
An orchestrator splits the task, several agents work at once, and their results are merged.
Best for tasks that break into independent partsHow it works
The Parallel Multi-Agent System pattern
One orchestrator decomposes the request into independent sub-tasks and dispatches them simultaneously. Each agent works without waiting for the others. When all have finished, an aggregator merges their outputs into a single response.
- 1
Orchestrator
Splits the request into independent sub-tasks.
- 2
Agents run in parallel
Each sub-task is handled at the same time by a dedicated agent.
- 3
Aggregator
Waits for all agents, then merges their outputs into one answer.
Why it matters
Wall-clock time is set by the slowest agent, not the sum of them, so independent work finishes far faster. It only helps when the sub-tasks genuinely do not depend on each other.
Playground
Run the parallel multi-agent system
Your Question
InputA question that breaks into independent parts
Orchestrator
AgentSplits the task and dispatches sub-agents at once
Search Agent
AgentGathers current specs and pricing
Comparison Agent
AgentBuilds a spec-by-spec comparison
Review Agent
AgentSummarises aggregate owner feedback
Aggregator
AggregatorMerges the three results into a recommendation