This section covers grouping several specialist agents together under one manager, so one conversation can be handled by whichever agent is best suited to it.
Overview
An AI Team has member agents — each good at something specific — and a Team Manager that reads each request and decides which member should handle it, based entirely on the instructions you write and each member's own description. Delegated work runs inside the same session as the manager, on its own sub-task thread, so the whole exchange shows up as one conversation in Sessions and Traces even though several agents contributed to it.
What's in This Section
Topic | Covers |
|---|---|
AI Teams | Creating a team, what sections it has, and how it relates to individual agents. |
Team Manager | The model, routing instructions, and tools that drive delegation. |
Team Agents | The member roster — adding, removing, and governing who's eligible. |
Team Orchestration | What happens at runtime — routing styles, hand-backs, and loop protection. |
When to Use a Team Instead of One Agent
Reach for a team when a job genuinely splits across distinct specialties — billing, support, and analytics, for example — rather than trying to cram every capability into one agent's Instructions and Tools.