I am requesting a true Three-Way Mode in Codex: one visible, continuous conversation between the user, Codex, and a subagent.
This would be a game changer for how I work. Today, if I want a genuine second opinion on Codex’s work, I have to copy the task, diff, context, and test results into another chat, then copy the feedback back again. That is slow, loses context, and makes me the courier between agents.
Instead, I want one shared room where:
- I can direct the work and make the final calls.
- Codex remains the primary implementation agent with normal repository, terminal, editing, and verification powers.
- A subagent acts as an independent, read-only reviewer by default: it can inspect Codex’s real changed files, diff, test output, and relevant tool evidence, then give direct feedback.
All three participants should be visibly named in the same timeline. I should be able to @mention Codex or the reviewer. Crucially, Codex and the reviewer should be able to address each other directly: the reviewer can point out a bug, missing requirement, weak assumption, regression risk, or missing test; Codex can respond, fix it, or explain the trade-off; and I can see the whole exchange and decide.
This is not a request for a hidden subagent, a generic group chat, or a consolidated summary after the fact. The value is the visible three-way working conversation and the independent third opinion while the real work is happening.
The permission model can stay safe and familiar: Codex retains normal approval boundaries, while the reviewer is read-only unless I explicitly authorise more.
Current subagents are already valuable for delegation. What is missing is a first-class collaboration surface that lets the human, primary agent, and reviewer genuinely work together in one conversation—without copy-and-paste.
Is this kind of Three-Way Mode on the roadmap, or something the Codex team would consider?