Approve the next step from a phone while the job runs elsewhere
As artificial intelligence grows capable of sustaining long, complex tasks across hours rather than moments, OpenAI has extended its Codex coding agent to mobile devices — allowing developers to remain present in their work without being physically tethered to a desk. The move reflects a broader shift in how humans and machines collaborate: not in bursts of instruction, but in ongoing relationships of oversight and judgment. More than four million weekly users now carry the ability to approve, redirect, and observe AI-driven development from wherever life takes them, a quiet but significant reordering of where work ends and the rest of life begins.
- AI coding tasks now run for hours and demand human judgment at unpredictable moments, making desk-bound supervision increasingly impractical for developers.
- OpenAI's mobile Codex preview lands on iOS and Android this week, letting developers monitor live terminal output, code diffs, and approval requests from their phones in real time.
- A secure relay layer keeps sensitive files and credentials locked on the originating machine while only decisions and information travel to and from the phone, preserving security without sacrificing mobility.
- Remote SSH moves to general availability, allowing Codex to plug directly into enterprise-managed environments with existing security policies and compute resources already in place.
- Enterprise teams gain programmatic access tokens, customizable workflow hooks, and HIPAA-compliant options, positioning OpenAI's coding agents as infrastructure-grade tools for regulated industries.
OpenAI has brought its Codex coding agent to mobile phones, allowing developers to supervise long-running programming tasks from iOS and Android devices while the actual work continues on laptops or remote servers. The preview, available this week across subscription tiers in supported regions, lets users monitor progress, review outputs, approve next steps, and even start new work — all from a phone.
The shift speaks to how AI coding tools have matured. Where they once handled quick, keyboard-driven requests, they now manage jobs that unfold over hours and occasionally require human judgment mid-task. With more than four million weekly Codex users, the ability to supervise from anywhere has become practical necessity rather than convenience.
The mobile experience surfaces live updates including screenshots, terminal output, code diffs, test results, and approval requests. Crucially, files, credentials, and permissions remain on the machine running Codex — the phone only receives information and sends back decisions, connected through a secure relay that keeps remote machines reachable without exposing them to the internet directly.
OpenAI also made Remote SSH generally available, letting Codex connect to managed development environments teams already rely on — complete with approved dependencies and security policies. A task begun on a desktop can be handed off seamlessly to mobile supervision through the same relay infrastructure.
For enterprise customers, new programmatic access tokens enable integration with CI pipelines and release workflows, while generally available hooks allow teams to scan prompts, run validators, log conversations, and tailor Codex behavior to specific repositories. HIPAA-compliant use is now supported for eligible ChatGPT Enterprise workspaces in local environments, opening the door for healthcare organizations.
Availability is tiered: Remote SSH and hooks work across all plans, programmatic tokens are limited to Enterprise and Business, and HIPAA compliance is restricted to eligible Enterprise workspaces. Windows mobile support has not yet launched. Together, these moves signal OpenAI's conviction that AI coding agents will take on increasingly complex, multi-step work — and that the humans guiding them will need to do so from anywhere.
OpenAI has brought its Codex coding agent to mobile phones, letting developers step away from their desks while still keeping watch over long-running programming tasks. The feature arrived this week in preview on both iOS and Android, available to users across all subscription tiers in supported regions. What this means in practice is that a developer can start a complex job on a laptop or remote server, then monitor its progress, review its work, and approve its next steps from a phone—all in real time.
The shift reflects how AI coding tools have evolved. A few years ago, these systems handled quick requests typed at a keyboard. Now they're being asked to manage jobs that unfold over hours, making decisions along the way that sometimes need human input. More than four million people use Codex each week, according to OpenAI, and as those tasks grow longer and more intricate, the ability to supervise from anywhere becomes practical rather than optional.
From the mobile app, a developer can see what's happening on a connected machine—which threads are active, what approvals are pending, what context the job needs. They can review outputs, approve commands, switch between different AI models, or start entirely new work. The system sends live updates to the phone that include screenshots, terminal output, code diffs, test results, and requests for human judgment. Importantly, the actual files, credentials, and permissions stay on the machine where Codex is running. The phone only receives information and sends back decisions. A secure relay layer keeps the remote machines reachable across devices without exposing them directly to the internet.
OpenAI outlined several practical scenarios. A developer investigating a bug while away from the office can let Codex inspect files and run tests in the development environment, then approve the next steps remotely if needed. During a lengthy code refactor, Codex might reach a decision point and ask the user to choose between two approaches—something that can now happen instantly on a phone rather than waiting until the developer returns to a desk. The app can also prepare briefings for customer support conversations by pulling together updates from messages, documents, and web tools. And when an idea strikes away from work, a developer can start a new thread from their phone.
Alongside the mobile launch, OpenAI made Remote SSH generally available. This feature lets Codex connect directly to managed remote environments that development teams already use—systems with approved dependencies, security policies, and compute resources already in place. The desktop app can detect hosts from a user's SSH configuration and create projects inside those remote machines the same way it would on a local device. Once connected, authorized ChatGPT devices can supervise that work through the same relay infrastructure, meaning a developer can begin a task on a desktop and continue watching it from a phone.
For business customers, OpenAI introduced new management and automation tools. Programmatic access tokens can now be issued from ChatGPT workspace settings, enabling continuous integration pipelines, release workflows, and internal automation. Hooks are now generally available, allowing teams to scan prompts for secrets, run validators, log conversations, create memories, or customize Codex behavior for specific repositories and directories. OpenAI also now supports HIPAA-compliant use of Codex in local environments for eligible ChatGPT Enterprise workspaces, aimed at healthcare organizations handling patient care and operational workflows.
Availability varies by tier. Remote SSH and Hooks work on all plans. Programmatic access tokens are limited to Enterprise and Business plans. HIPAA compliance is restricted to eligible Enterprise workspaces using Codex in local environments. Support for connecting phones to Codex on Windows has not yet launched. The mobile preview itself is available on iOS and Android for users on Free, Go, and paid plans in supported regions. The rollout signals OpenAI's confidence that AI coding agents will increasingly handle complex, multi-step development work—and that developers will need to supervise those agents from anywhere.
Citações Notáveis
As coding agents take on longer tasks, users increasingly need to step in at small but important moments, such as answering questions, reviewing outputs or approving the next step.— OpenAI
A Conversa do Hearth Outra perspectiva sobre a história
Why does a developer need to approve coding tasks from a phone? Can't the AI just finish the job?
Because these aren't simple tasks anymore. Codex might be refactoring thousands of lines of code and hit a fork in the road where two approaches are equally valid—but one fits your project better. You don't want to wait eight hours to get back to your desk to make that call. The job stalls. With mobile approval, you decide in thirty seconds from wherever you are.
So the actual code stays on the server, and the phone just sees what's happening?
Exactly. Your credentials, your files, your local setup—all of that stays secure on the machine where Codex is running. The phone gets screenshots and diffs and test results. You send back yes or no or a new instruction. It's a clean separation.
Four million weekly users is a lot. Are most of them using this mobile feature?
It's in preview, so it's new. But the fact that OpenAI is building it tells you something about where they think this is headed. They're not expecting developers to use Codex for five-minute tasks anymore. They're expecting longer jobs that need human judgment at key moments.
What about teams? Is this just for solo developers working from home?
No. They added Remote SSH so teams using centrally managed development environments can connect Codex to those shared systems. And they added programmatic tokens so enterprises can wire Codex into their CI/CD pipelines and automation workflows. This is enterprise-grade infrastructure now.
HIPAA compliance—that's healthcare, right?
Yes. It means healthcare organizations can use Codex for patient-facing systems and operational workflows without violating privacy regulations. That's a significant unlock for an entire industry.
What's the catch?
Windows phone support isn't here yet. And some features are locked to Enterprise and Business plans. But the core mobile preview is available to anyone on a paid plan. The infrastructure is there. It's just a matter of which features OpenAI wants to keep for premium customers.