Multi-agent

Hierarchical teams

Managers of managers — layers of agents for large scope.

N agentsAdvanced

How it works

1A top agent splits the goal into domains and assigns a lead per domain.
2Each lead further splits its domain among worker agents.
3Workers report up; leads synthesize their domain.
4The top agent assembles the whole. Define clean interfaces so layers don't leak context.

Use it when

Very large tasks that need division into domains, each with its own sub-team — a "research org" with section leads and researchers.

Reach for something else when

Most tasks. Layers add latency, cost, and failure points — use the flattest structure that works.

Where you stay in the loop

You design the org and its interfaces; the layers delegate and report. The more layers, the more context can leak — audit what each layer passed up versus what it actually meant.

In the wild

A report-writing org: an editor-in-chief assigns sections to section leads, who assign sub-topics to researchers.

Hand this to your agent

Design a hierarchical agent team for my goal.

Output the org chart: a top coordinator, 2–4 domain leads, and the worker
roles under each — one line on what each owns and what it passes upward.

Then run it top-down: the coordinator delegates, leads sub-delegate, workers
produce, and results roll back up into a final synthesis.

Goal: <...>

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