Claude Code Agent Teams let one Claude Code session act as a lead that assigns tasks to teammate sessions, each with its own context window; teammates claim work from a shared list, message each other and report back. An agent room lets several people send work to one owner’s agent, under permissions the owner sets, with dangerous actions pausing for approval. Teams multiply agents for one person. Rooms multiply people for one agent.
They are sometimes confused because both involve “more than one” and both keep everything on your machine. They answer different questions and they compose.
One person, many agents on the left; many people, one agent on the right.
What Agent Teams solve
Throughput and separation of concerns within a single user’s task. A lead breaks a refactor into pieces; three teammates work on three modules in parallel; the lead integrates. You can talk to a teammate directly. Everything runs as you, on your machine, under your permission mode. It is orchestration.
What a room solves
Access and accountability across people. A co-founder, a designer and a client can each ask the agent for work, each under a permission mode or a standing grant you chose, with pushes, deletions, spending and credential access stopping for you, and a record of who asked and who allowed. It is governance.
Side by side
How they stack
An agent in a room is a Claude Code session. Nothing stops that session from running an agent team internally when a request is large. The room sees one request, one timeline and one result; inside, the lead and its teammates did the work. Approvals still apply: if any teammate reaches for git push origin main on a request from someone else, that pauses for the owner.
So the sensible mental model is: rooms decide who may ask and what must pause; teams decide how the agent gets it done.
When you need which
- You alone, big task, want speed: Agent Teams.
- Other people need to ask your agent: a room.
- Other people ask, and their tasks are big: both.
Frequently asked questions
Can a room member start an agent team? They send a request. Whether the agent uses a team to fulfil it is up to the agent and the owner’s configuration. The member sees the request’s timeline either way.
Do teammates get their own permissions in a room? No. Teammates are internal to the agent’s run. Permissions in a room attach to people, not to sub-agents.
Does a room need Agent Teams? No. Most requests in the first month of our beta took under a minute and ran as a single session. See state of shared agents, September 2026.
Multiply your agents with a team. Multiply who can ask with a room. Give your agent one.
Published