What changed
OpenAI announced on May 14 that Codex is now available in preview inside the ChatGPT mobile app, so developers can monitor and steer long-running coding tasks from their phones while the actual work continues on a laptop, devbox, or other managed environment. The same announcement also packaged several workflow upgrades around Codex: Remote SSH is now generally available, Hooks are generally available, and programmatic access tokens can now be issued from ChatGPT workspace settings for trusted automation. Source: OpenAI product announcement.
Why this matters
The interesting part is not just “AI on mobile.” It is the shift toward asynchronous software work: you can start a longer task on your main machine, review questions or approvals from your phone, and keep the task moving without waiting to get back to your desk. That makes coding agents more practical for real-world work like bug triage, refactors, support follow-up, and CI-style automation.
For teams, the remote setup matters even more. OpenAI’s Codex docs say remote projects can run directly against an SSH host, so the code, dependencies, credentials, and security policies stay in the environment where work already happens. That is a better fit for serious development than moving code into a separate hosted sandbox. Source: Remote connections docs.
The product details worth watching
- The mobile experience is not just a notification layer. OpenAI says the app can load live state from your Codex environment so you can review outputs, approvals, diffs, test results, and active threads from your phone.
- Remote SSH support means Codex can run directly on a devbox or managed remote machine instead of only a local laptop. OpenAI’s docs also explicitly recommend keeping normal SSH security expectations in place rather than exposing an app server publicly.
- Hooks give teams a deterministic way to add policy and workflow logic around Codex runs. OpenAI documents examples like blocking accidentally pasted API keys, logging conversations, creating persistent memories, and running validation checks at turn boundaries.
- Programmatic access tokens make the update more useful for ops-heavy teams because they let trusted automation run Codex non-interactively with a ChatGPT workspace identity, instead of requiring a human browser login.
Bottom line
This looks like a meaningful product step for developer-facing agents: less “single chat window,” more continuous collaboration across devices and real environments. If you already use coding agents, the practical takeaway is straightforward: Codex is moving closer to the shape of an always-on teammate that you can start on desktop, unblock from mobile, and wire into automation with guardrails.