> ## Documentation Index
> Fetch the complete documentation index at: https://agentic.proxify.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Teams

> When and how to use Agent Teams — the decision framework, delegate mode, and practical advice.

<Note>**You'll learn:** when Agent Teams justify their \~15x cost, how delegate mode works, and the practical failure modes to watch for.</Note>

<Tip>**Official reference** — The [official Agent Teams docs](https://docs.anthropic.com/en/docs/claude-code/agent-teams) cover setup, architecture, keyboard shortcuts, display modes, and limitations. This page adds the decision framework and practical advice.</Tip>

<Warning>Agent Teams is **experimental**. Expect rough edges — no session resumption, occasional task status lag, and slow shutdown.</Warning>

## When to Use Agent Teams

Use this decision framework:

| Situation                                                   | Tool                                             | Why                                       |
| ----------------------------------------------------------- | ------------------------------------------------ | ----------------------------------------- |
| Fully independent work (migrations, convention enforcement) | [`/batch`](parallel-execution#the-batch-command) | No coordination overhead, cheapest option |
| Tasks that report results back (research, analysis)         | [Subagents](subagent-patterns/)                  | Parent orchestrates, \~4-7x cost          |
| Tasks that need to discuss and coordinate                   | **Agent Teams**                                  | Shared task list + messaging, \~15x cost  |

Agent Teams justifies its \~15x token cost only when agents genuinely need to share findings, challenge each other's assumptions, or self-coordinate through a shared task list.

## Delegate Mode Is Essential

Press **Shift+Tab** to cycle into delegate mode, which restricts the lead to coordination-only tools.

<Warning>Without delegate mode, the lead frequently implements tasks itself instead of delegating. Practitioners consistently flag this as the most common Agent Teams failure mode.</Warning>

In delegate mode, the lead can only:

* Create and manage tasks
* Send messages to teammates
* Review and approve plans
* Synthesize results

## A Typical Session

```
You: "Create a team for the auth refactor"
Lead: Creates team, spawns 3 teammates:
  - backend-auth (implements API changes)
  - frontend-auth (updates UI components)
  - auth-tests (writes integration tests, blocked by backend-auth)

You: Enable delegate mode (Shift+Tab)
Lead: Creates tasks, assigns to teammates
Teammates: Work independently, message each other when interfaces change
You: Monitor with Ctrl+T, review output as teammates complete
Lead: Synthesizes results, resolves conflicts
```

## Practical Advice

* **Sweet spot: 3-5 teammates.** Fewer underutilizes parallelism; more creates coordination overhead exceeding the throughput gain.
* **Use plan approval** for high-stakes work — teammates submit plans before implementing, the lead reviews.
* **Prepare for no session resumption.** If you need to `/resume`, tell the lead to spawn replacement teammates.
* **Nudge on task status lag.** Teammates sometimes forget to mark tasks complete — check with Ctrl+T.

***