OpenAI announced today that Codex remote control is now in the ChatGPT mobile app on iOS and Android, in preview, available to all subscription tiers. The mobile surface lets users monitor live Codex environments across devices and — quoting the announcement — "work across all of your threads, review outputs, approve commands, change models, or start something new." The change ships into the existing ChatGPT app rather than as a standalone Codex app. No release date for general availability was given; preview is the current status.
The actual capability shift here is the asynchronous-agent + mobile-checkpoint working mode. Until now, running a long Codex coding session meant sitting at the machine to approve tool calls and review intermediate outputs. With mobile remote control, the session continues on the server, sends approval prompts to your phone, and you stay in the loop while moving through your day. That working pattern is not new in the industry — Anthropic shipped Remote Control for Claude Code in February of this year, and OpenAI's update is the competitive-parity move. The OpenAI sequence over the last six weeks has been steady: April brought desktop background-execution, then the Chrome extension for browser automation, and now mobile. The arc is the same arc: pulling the agent off the developer's workstation and putting checkpoint approvals on whatever screen is closest.
For builders, the meaningful comparison is not "did OpenAI catch Anthropic" but "what does asynchronous agent supervision change about how you structure work." Long-running agent sessions only make sense if the supervision overhead drops below the cost of just doing the task yourself. Mobile approval flow lowers that overhead by making the agent's blocking moments — "should I run this command, modify this file, hit this API" — interrupt-driven rather than tied to where you happen to be sitting. The trap to watch for: too-easy mobile approvals turn into rubber-stamping, and rubber-stamping turns into agents shipping things you would have stopped on a desktop review. The right calibration is mobile for fast read-only checks and short approvals, desktop for any change with non-trivial blast radius.
The honest caveats: this is preview, not general availability; the announcement does not specify which Codex models are exposed on mobile or whether there are mobile-specific rate limits; no details on supported coding tasks beyond the workflow-management framing, and nothing on the API or MCP side. If you ship development workflows on Codex, the practical thing to do is enroll in the preview, run a real session through the mobile approval surface, and verify what blast-radius operations actually require your input vs proceeding silently. The capability is the right one to ship; the deployment-time question is always whether the approval gating matches the risk model of the tasks you're running.