You can now control OpenAI’s Codex — the agentic coding assistant that writes code and operates desktop applications — directly from your phone. The feature, rolling out today as a preview on iOS and Android, represents a fundamental shift in how developers interact with AI coding agents: the agent is no longer tethered to the machine where it runs.
Announced on OpenAI’s blog and quickly shooting to #5 on Hacker News with 309 points, the move is a direct response to the surging popularity of Anthropic’s Claude Code, which has set the standard for agentic coding workflows — as documented in our complete guide to AI agents over the past six months. OpenAI is now fighting back — not by matching Claude Code feature-for-feature, but by making Codex available in the one place no other coding agent has gone: your pocket.
What Actually Changes
Before today, Codex was a desktop-only experience. You installed it on your machine, it had access to your files, terminal, and browser — and you interacted with it through a local interface. Useful, but tethered.
The mobile integration flips that model — a shift reflecting the trends tracked in our state of AI agents May 2026 overview. From the ChatGPT app on your phone, you can:
- Start new tasks — Tell Codex what to build or fix, and it begins working on your desktop machine
- Review outputs in real time — Screenshots, terminal output, and diffs stream back to your phone as Codex works
- Approve or reject commands — Codex pauses at decision points and sends a prompt to your phone for approval before executing potentially destructive actions
- Monitor multiple threads — Keep an eye on several Codex sessions simultaneously, switching between them from the app
- Receive notifications — Get pinged when Codex finishes a task, hits an error, or needs your input
The key architectural detail: your files, credentials, permissions, and local setup never leave the desktop machine. The mobile app is a remote control, not a cloud sync. Chunks of state — screenshots, terminal excerpts, diff summaries — flow to your phone, but the actual codebase stays where it belongs.
“Your files, credentials, permissions, and local setup stay on the machine where Codex is operating, while updates flow back to your phone in real time, including screenshots, terminal output, diffs, and more.” — OpenAI announcement
Availability and Pricing
The feature is rolling out across all ChatGPT plans, including the free tier and the more affordable Go plan. This is notably more accessible than Claude Code, which requires a paid Anthropic subscription and remains desktop-only.
| Feature | Codex (Mobile) | Claude Code (Desktop only) |
|---|---|---|
| Mobile remote control | ✅ Yes | ❌ No |
| Desktop execution | ✅ Yes | ✅ Yes |
| Free tier access | ✅ Yes | ❌ No |
| Real-time output streaming | ✅ Yes | ✅ Terminal only |
| Multi-session management | ✅ Yes | ❌ No |
| Permission gates | ✅ Yes | ✅ Yes |
Why This Matters for the Agent Ecosystem
This launch signals something larger than a mobile port. It’s a re-architecting of the human-agent interface.
The Desktop Lock Is Breaking
Every major coding agent today — Claude Code, Codex, Cursor, Windsurf — operates on the assumption that the human sits at the same machine as the agent. The interaction is synchronous: you type, the agent responds, you review, you iterate. All in one terminal window on one screen.
Codex mobile breaks that assumption. The agent runs on one machine; the human interacts from another. This is the difference between a tool (something you use) and a worker (something you delegate to). When you can tell your agent to “finish that PR while I’m in the coffee line,” the relationship changes. The agent becomes asynchronous, background, persistent — more like a team member than an IDE plugin.
The Approval Gate Model Goes Mobile
The mobile integration forces a design pattern that every agent platform will eventually need: approval gates over push notifications. Codex doesn’t just run wild — it pauses at decision points (file writes, git commands, destructive operations) and pushes an approval request to your phone. You review the context — a diff, a screenshot, a terminal excerpt — and tap approve or reject.
This pattern mirrors how Claude for Small Business (launched by Anthropic just yesterday) handles approval gates for financial workflows. The industry is converging on a model where agents act autonomously within a bounded scope, but punt high-risk decisions to human supervisors — wherever those supervisors happen to be.
The Competitive Landscape Shifts
The timing is telling. Yesterday, Anthropic launched Claude for Small Business with 15 pre-built agentic workflows for QuickBooks, PayPal, and HubSpot. Today, OpenAI drops Codex mobile. The two announcements bookend a clear strategic divergence:
- Anthropic is going vertical — deep integrations into specific business tools, targeting non-technical users
- OpenAI is going horizontal — making its existing coding agent ubiquitous across devices, targeting developers wherever they are
Both strategies are valid. Both recognize the same truth: the next battle in AI agents isn’t about model quality — it’s about accessibility and integration.
What’s Still Missing
The mobile preview is useful, but it has limitations worth noting:
- Latency — The two-machine relay introduces a perceptible delay, especially for large diffs or high-resolution screenshots
- No offline mode — Both the desktop and mobile endpoints need internet connectivity
- Mobile-only output viewing — You can review and approve, but detailed code editing on a phone screen is impractical
- Single desktop binding — Each mobile session connects to one desktop Codex instance; multi-machine orchestration isn’t supported yet
These are early-days limitations. The architecture — desktop execution with mobile control — opens up possibilities that no other agent platform has explored: multi-modal supervision, location-agnostic approval chains, and eventually, delegation hierarchies where one human supervises a fleet of agents from a single phone.
The Bigger Picture
Codex in the ChatGPT mobile app is more than a feature update. It’s the first credible answer to a question the agent ecosystem has been dancing around: what happens when you’re not at your desk?
The answer, as of today, is that your agent keeps working. You approve merges from the train. You review diffs from the café. You start a refactor before walking into a meeting and come back to a finished pull request.
This is the direction the industry is heading: agents that don’t wait for you. The question is whether Claude Code, Cursor, and the rest can match this mobility before OpenAI establishes mobile-first agent supervision as the new default.
Codex mobile is rolling out now on iOS and Android. Available across all ChatGPT plans. Follow the discussion on Hacker News or read the full announcement on OpenAI’s blog. Coverage also from The Verge.