Multi-agent

Orchestrator–workers

A lead agent breaks down the job and delegates to workers.

N agentsCore

How it works

1Orchestrator analyzes the task and dynamically splits it into subtasks.
2Spins up a worker per subtask with a focused brief.
3Workers run (often in parallel) and return their results.
4Orchestrator integrates the results into the final output.

Use it when

Tasks whose subtasks aren't known until you see the input, and benefit from parallel specialized work the lead then synthesizes.

Reach for something else when

Subtasks are fixed and simple — a plain chain or parallel split is cheaper and more predictable.

Where you stay in the loop

The orchestrator divides the work and synthesizes the results; you judge whether the synthesis is coherent and the division made sense. Conflicts between workers are where your call matters most.

In the wild

A coding agent decides which files need changes, assigns each file to a worker, then merges and reconciles their edits.

Hand this to your agent

Act as an orchestrator agent.

For my task:
  1. Break it into independent subtasks. For each, describe the worker brief:
     role, input, expected output.
  2. I'll run each worker (or you may simulate them one at a time).
  3. Collect the worker outputs I paste back and synthesize them into one
     coherent result, resolving any conflicts between workers.

Task: <...>

Replace the <…> placeholders, paste it into your agent, and it'll scaffold the workflow with you.