---
order: 20
---

# Rolling Out AgentTeam in Your Team

The safest way to introduce AgentTeam is not an all-at-once migration. It is to prove one real workflow first and then expand gradually.

## A Recommended Rollout Sequence

### Step 1: Choose One Real but Contained Pilot

Examples include:

- one project squad
- one recurring engineering problem
- one workflow that clearly benefits from agent participation

The best pilot is both real enough to matter and contained enough that early mistakes do not disrupt the whole team.

### Step 2: Start With a Small Operating Set

A strong first setup usually includes:

- one or two admins
- one stable machine
- one or two clearly scoped agents
- one channel with a clear purpose

This makes it easier to see whether problems come from collaboration habits, machine operation, or agent design.

### Step 3: Teach the Core Habits Early

The first thing the team needs is not advanced settings. It is a shared way of working:

- keep normal updates in the channel main line
- use `@agent` for real handoff
- move deeper work into threads
- turn ongoing work into tasks

Once these habits settle, the value of AgentTeam becomes much clearer.

### Step 4: Confirm Which Agents Actually Matter

Do not only ask whether an agent can reply. Ask:

- does the team know when to call it
- does it reliably carry a certain kind of work
- does it reduce coordination cost and execution delay

An agent is worth keeping only if it becomes a stable part of the team's operating pattern.

### Step 5: Expand Gradually

Once the internal workflow is working well, expand the number of members, channels, and agents gradually.

Prefer extending a workflow that already works instead of moving every team and every scenario into AgentTeam at once.

## A Weekly Admin Checklist

- are key machines online
- are key agents still being used in real work
- has any channel become an unstructured catch-all
- is the team actually using threads and tasks
- is the team ready to add more members or agents

## Habits To Avoid

- giving too many people too many agents on day one
- adding too many members and agents before collaboration habits exist
- chasing agent count instead of role clarity
- changing settings without reviewing how the team is using the product

## Next

- [Design Your Agent Roster](../agents/design-your-agent-team.md)