Skip to Content
Docs are being rebuilt — start at Introduction → How it works.
GuidesBy departmentCross-functional & power-userMulti-agent research team

Multi-agent research team

You’ll build a coordinator agent that delegates parts of a research task to specialist agents and pauses for human approval before any consequential action.

The plan

  1. Build the specialist agents
  2. Build the coordinator
  3. Wire up delegation
  4. Require approval for actions

Step 1 — Build the specialists

In Agents, create a few narrow agents — for example a web researcher, a data analyst, and a writer — each with focused system instructions. Give each only the capabilities it needs.

Step 2 — Build the coordinator

Create a coordinator agent whose instructions describe when to delegate and to whom. Have specialists end with a short <final_answer> block so the coordinator receives clean results — see Agent-to-agent calls.

Step 3 — Wire up delegation

Follow the hierarchical pattern: the coordinator calls each specialist, reads back the result, and assembles the final output. See Multi-agent collaboration for the full picture.

Step 4 — Require approval

When the team can take real actions (sending, posting, writing back), turn on Require Approval in the Capabilities tab so a human signs off before tools run. Test from the builder preview.

Where to go next

Cross-functional & power-userVoice receptionist