Codex Cloud and GitHub Copilot Cloud Agent: Where Delegated Code Runs
A practical comparison of setup, execution environments, review workflows and documented constraints for two cloud coding agents.
Delegated coding is most useful when a task can run independently while you keep control of the final changes. Codex Cloud and GitHub Copilot cloud agent both move coding work into a remote environment, but their setup and review paths differ. Choose by where your team already manages repositories and how much environment preparation you want the service to handle.
Codex Cloud starts with a prepared project environment
In ChatGPT web or desktop, choose Work in → Cloud and select an existing environment or create one. For a new setup, choose GitHub repositories and connect GitHub if prompted. Codex inspects the project, installs dependencies and tools, and tests the workflow; you supply missing access or information. Review setup and test results, save the changes, and publish the environment before starting a task.
The published environment can be reused across tasks. Each task receives its own workspace, and tasks can continue while your computer is asleep. OpenAI's documentation says you can connect package registries, APIs and private services by configuring network access and supplying credentials in environment settings. That makes setup review important: grant only the service access the project requires.
Copilot works in a GitHub Actions-powered environment
GitHub describes Copilot cloud agent as working in its own ephemeral development environment powered by GitHub Actions. It can explore repository code, edit files, run automated tests and linters, and make changes on a branch. Work can start from GitHub.com, an issue, Visual Studio Code, or a pull-request comment; GitHub also documents scheduled or event-triggered automations.
For Business and Enterprise subscribers, an administrator must enable the relevant policy. Repository owners can opt out some or all repositories, and GitHub says the agent is available in repositories stored on GitHub except managed-user-account repositories and those where it is disabled. GitHub's page identifies paid Copilot plans as eligible, but these documents do not establish a comparative price or every account's access; check your own organization policy and current account settings.
Review remains a human decision
Codex presents changed files and test results for review, allows follow-up requests, and offers commit or pull-request creation when you are ready. Copilot's flow is branch-centered: inspect its changes, iterate, then create a pull request. GitHub says its cloud-agent actions are visible in commits and logs. In either case, remote execution does not make generated code approved: inspect the diff, run the checks that matter to your project, and verify the result before merging.
Pick by setup and workflow
Codex is a fit when you want to publish and reuse a project-specific environment, including configured tools and service access, then dispatch separate tasks. Copilot cloud agent fits a GitHub-centered workflow where issues, branches and pull requests are already the natural handoff points. Both support asynchronous delegation; this comparison does not establish that either is faster or produces better code.
There are documented boundaries. GitHub says deep research, planning and iteration before opening a pull request are available on GitHub.com, with public-preview support in Microsoft Teams and Slack; other integrations such as Azure Boards, Jira or Linear support direct pull-request creation only. OpenAI's cloud guide points readers to separate environment documentation for networking, secrets and saved state, so review those controls before connecting private services. Start with a bounded, testable task and keep final review in your normal engineering process.
Try the related loot
GoDrop: Give Coding Agents a Shareable File URL
