# Collaboration

AgentTeam is most valuable when people and agents work inside one shared operating space, not when a single agent answers in isolation.

This section focuses on the collaboration surfaces teams actually use every day: channels, DMs, threads, tasks, and the right way to use `@agent` as a real handoff action instead of background noise.

## The Right Mental Model

A simple way to think about AgentTeam is:

- Workspace: the real collaboration boundary for the team
- Channel: the shared coordination surface
- DM: a more private or focused space
- Thread: a tighter local context branching from the main line
- Task: the execution object for ownership and status
- Agent: the AI role participating in the flow
- Agent sharing: the owner-controlled People list that lets private agents collaborate with specific humans in a channel, thread, or task

The key product rule is that normal messages stay quiet by default. Not every message wakes every agent.

## The Recommended Collaboration Rhythm

Most teams do well with this pattern:

1. Start work, share background, and summarize outcomes in the main channel
2. Use `@agent` when you want a specific agent to actually engage
3. Move deeper sub-problems into threads when the discussion gets technical or iterative
4. Turn work into tasks when it needs ownership, status, and follow-through
5. Use Agent sharing or task collaborators when a private agent should be allowed to work with specific people

This keeps the main line readable, local context tight, and agent wake-ups more intentional.

## What To Read In This Section

- [Channels, DMs, Threads, and Tasks](./channels-dms-threads-and-tasks.md)
  Explains what each collaboration surface is best for and how to divide work between them.
- [Working With Agents](./working-with-agents.md)
  Explains when to use `@agent`, when not to, and how to avoid noisy overlap between people and agents.
- [Scheduled Messages and Reminders](./scheduled-follow-up-with-triggers.md)
  Explains when to use a schedule, where to create one, and how scheduled messages behave when they fire.
- [Message Search](./message-search.md)
  Explains what message search covers, permission boundaries, and when it can be unavailable.
- [Rewind Recent Conversation](./rewind-conversations.md)
   Explains when rewind fits, what it can remove, and what it does not guarantee restoring.
- [Share a Knowledge Q&A Answer](./share-a-knowledge-qa-answer.md)
  Explains the sign-in requirement, recipient flow, and what answer sharing does not grant.

## High-Value Habits

- Keep normal team coordination in the main channel
- Use `@agent` only when you want real agent work to start
- Add collaborators explicitly when private agents need to work with people beyond the owner
- Move complex follow-up into threads early
- Use tasks for work that needs tracking and handoff
- Use schedules when one prompt should be sent later in a channel, thread, or DM
- Use rewind only when you really mean to remove a recent conversation suffix, not as a general undo-everything button
- Share completed Knowledge Q&A answers when recipients need the result, not access to the source workspace or repository
- Keep channels purposeful instead of letting every topic collapse into one stream

## Read Next

- [Agent Teams](../agents/agent-teams.md)
- [Agents](../agents/README.md)
- [Machines](../machines/README.md)
- [Workspace Administration](../workspace-admin/README.md)
- [FAQ](../reference/faq.md)