Skip to content

Operate an Agent Together

An agent session is easier to trust when more than one person can see it. Pods make an agent session a shared place: everyone you invite attaches to the same live terminal — same screen, same scrollback, same agent — instead of trading screenshots or screen-shares.

Invite your teammates

  1. Launch a pod and start your agent — for example claude in a Claude Code pod — in the pod's main terminal (the shared one).
  2. Share the pod with members of your organization, choosing a role for each:
  3. Read-write — full participants: they can drive the shared terminal (see below) and open terminal tabs of their own.
  4. Read-only — they see the same live terminal but can't type.
  5. Collaborators open the pod from their own workspace and land in the same main terminal. Everyone sees the agent's output as it happens.

One driver at a time. The shared terminal has exactly one keyboard at any moment. The presence panel shows everyone in the session and who is currently the Driver; anyone else with read-write access can click request control, and the driver hands the keyboard over. No stepped-on keystrokes, no two people typing into the agent at once — and a clear record of who is driving.

Warning

A read-write collaborator can take the keyboard to your shell, with everything your pod can reach. Share deliberately, and give read-only access to anyone who only needs to observe.

Patterns that work

  • Driver and observers. One person drives the agent; reviewers watch live in read-only and speak up (in your usual chat or in the room) when the agent needs steering. Useful for risky operations — migrations, production changes — where four eyes are policy.
  • Follow-the-sun handoff. The agent keeps working in the pod's main terminal after you disconnect — so a teammate in the next time zone opens the same pod, takes the keyboard, and picks up mid-session. With the Claude Code pod, the agent's memory and login persist in the pod too, so the agent also remembers where things stand.
  • Teach by doing. Onboard someone by letting them watch a real agent session read-only, then flip them to read-write and give them control when it's their turn to try.
  • Parallel lanes. Read-write collaborators can open their own terminal tabs (up to 10 per pod) — keep the agent driving in the shared terminal while a teammate inspects logs or runs tests in a tab of their own, all on the same machine and workspace.

Why this beats screen-sharing

  • Nobody's laptop has to stay online — the session lives in the pod.
  • Joiners get a real terminal, not pixels: they watch it live, copy from it, and (with read-write) can take the keyboard and act.
  • The workspace, the repository, and the agent's context are all in one durable place (/workspace), so "come look at this" never turns into "let me push a branch first."

Access is managed by the pod's owner, and organization admins can see and stop any pod in the org from the Control Center.