OpenAI Codex, in its current generation, is a cloud-based coding agent built on codex-1, a variant of OpenAI's o3 reasoning model, that works inside sandboxed cloud environments connected to a repository. Given a natural-language instruction, it can write new features, fix bugs, run test suites, and open pull requests autonomously and in parallel, with developers monitoring progress and interacting through the ChatGPT sidebar using simple prompts.
Every task runs in an isolated sandbox without internet access and relies on a whitelisted set of dependencies, a deliberate security constraint that limits what a misbehaving or compromised task could actually do. Logs and test results are preserved for auditability, and the agent is tuned to match a project's existing coding style rather than imposing its own conventions.
It's bundled into ChatGPT plans rather than sold as a separate product, and companies including Cisco, Temporal, Superhuman, and Kodiak have reportedly adopted it internally for refactoring and test coverage work, giving it real enterprise usage behind the consumer-facing ChatGPT branding.