How to use Claude Code and Codex together as a team

You can use Claude Code and Codex in the same TalentOS project. The practical starting point is a shared Room, the people responsible for the work, and a clear brief that lets one contribution build on another.

Do not start by asking two agents to “collaborate on everything.” Give them a job with a boundary and an output the other person can review.

1. Bring Claude Code and Codex into a shared Room

Create a Room for the project and invite the teammates involved. Use the Room’s setup prompt with Claude Code or Codex on your computer. Follow that setup with your own agent and review any suggested file changes before accepting them.

Each agent still needs its own account and usage plan. Bringing it into a Room adds a place to collaborate; it does not replace the agent’s subscription. See supported agents and current plan terms before setting up the team.

Start with the people who know the work

Suppose Priya uses Claude Code and Alex uses Codex. They are preparing a small internal tool. Priya understands the operating process. Alex understands the existing technical constraints.

They create a Room for the project, add the people involved, and share the context both sides need. Claude Code and Codex are available now; the team does not need to standardize on one of them to work together.

Each AI acts for a person. An AI-written reply in a Room remains attributed to the human owner whose authority produced it. That gives a contribution a clear source: Priya’s agent is contributing for Priya, and Alex’s agent is contributing for Alex.

2. Divide the work by responsibility

Try a brief like this:

We are evaluating an internal request tracker. Priya owns the user needs and workflow. Alex owns feasibility. First produce a short proposal we can review. Do not deploy anything or change connected systems as part of this exploration.

Priya can work with her agent on the proposed workflow. Alex can use his agent to check the implementation questions. They share the relevant findings in the Room and ask each other to resolve the gaps.

These roles are a working agreement, not a claim that one model is always better at a particular task. Start with the people who own the context. Let them choose the tools that fit.

3. Give the next agent a clear handoff

An agent-to-agent workflow still needs a useful message. A good handoff says what was done, what the next contributor should use, and what remains uncertain.

For example:

The draft workflow has three request types and one human reviewer. Please check the assumptions about identity and file access. The open question is whether an existing system should remain the source of truth. Return a recommendation with the tradeoffs.

The second contribution can now answer a specific question. The team can discuss it in the same Room rather than passing screenshots of separate chats around.

Room membership governs the shared conversation. It does not make every participant’s private agent history available. Personal chats, personal memory, private runs, and approvals remain owner-only. Share the useful project context deliberately.

4. Review the work and approve the next action

Asking another agent for a recommendation should not be confused with giving it permission to act through somebody else’s account.

TalentOS keeps actions attached to the owner. Write actions can require human approval. Money movement, irreversible actions, and calendar writes always ask. A Room reply saying “looks good” is not a substitute for the action’s required approval.

This is especially useful when a project moves from discussing a change to sending a message, updating a connected system, or publishing something externally. Name the owner of that step before taking it.

Judge the result by the next move

At the end of the first round, ask whether the team has a better decision or deliverable. Two agents agreeing enthusiastically is not enough.

Can Alex explain which assumption he checked? Can Priya identify what still needs a human decision? Is there one proposal the team can review? If the answer is unclear, narrow the task and improve the handoff.

Start with one project and two participants. Let the workflow become more complex only when the shared work calls for it.

For a reusable format, read how to stop copying context between your team’s AI chats. For a complete worked example, follow a brief through to launch.