News
The Return of Codex AI — as an Agent
OpenAI's "Codex" AI model is back, in a new form from the 2021 offering that powered the original GitHub Copilot and kickstarted the GenAI craze.
Rather, the new Codex, now in research preview, is described as "A cloud-based software engineering agent that can work on many tasks in parallel, powered by codex-1. Available to ChatGPT Pro, Team, and Enterprise users today, and Plus users soon."
Back in 2021, the company said, "We've created an improved version of OpenAI Codex, our AI system that translates natural language to code, and we are releasing it through our API in private beta starting today."
That post has now been updated with:
Update on May 16, 2025: We launched Codex, a cloud-based software engineering agent that can work on many tasks in parallel. Learn more
Update on April 16, 2025: We launched Codex CLI, our new open source local coding agent. To get started, please visit https://github.com/openai/codex
The OpenAI Codex models were deprecated in March 2023. To learn about our latest coding models, please visit our docs.
The new Codex is powered by codex-1, a version of OpenAI o3 optimized for software engineering. It can perform tasks such as writing features, answering questions about your codebase, fixing bugs, and proposing pull requests for review. Each task runs in its own cloud sandbox environment, preloaded with your repository.
[Click on image for larger view.] Explain This Codebase to Me (source: OpenAI).
Highlights of the announcement include:
- Task Automation: Codex can write new features, answer questions about your code, fix bugs, and suggest pull requests.
- Cloud Sandbox Environment: Each task runs in an isolated cloud environment that is preloaded with your code repository for secure operation.
- Integration with ChatGPT: Codex is being integrated directly into ChatGPT for easy access.
- Availability: Currently rolling out to ChatGPT Pro, Enterprise, and Team users, with support for Plus and Edu users coming soon.
- Verifiable Evidence: Codex provides citations of terminal logs and test outputs, allowing users to trace its actions.
- Configurable Environment: Users can configure the Codex environment to closely match their local development setup.
- Multilingual Capabilities: Codex supports over a dozen programming languages, with strong performance in Python.
- API Release: The Codex API was launched in private beta to invite developers and businesses to integrate and experiment.
- Applications Beyond Code: The model could also translate between spoken language and commands, opening possibilities for voice-controlled software tools.
- Training Data: Codex was trained on a curated dataset including billions of lines of public source code available on GitHub.
- Use Cases: Demonstrated applications included building simple websites, data visualizations, games, and command-line utilities using natural language instructions.
- Commitment to Safety: OpenAI acknowledged the risks (e.g., bias, security vulnerabilities) and committed to ongoing testing and refinement.
Developers access Codex through the ChatGPT sidebar. To initiate a coding task, a user types a prompt and clicks "Code." For questions about code, you click "Ask." Each task runs in its own isolated environment with your codebase preloaded. Codex can read and edit files and execute commands like tests and linters. Task completion ranges from 1 to 30 minutes, and you can follow its progress live. Once done, Codex commits changes within its environment and provides verifiable evidence (terminal logs, test outputs). You can then review the changes, request revisions, create a pull request on GitHub, or integrate them locally. You can also configure the Codex environment to mirror your own development setup.
Going forward, OpenAI envisions a future where developers focus on the work they care most about while delegating the rest to intelligent coding agents like Codex. To support this, they're building a suite of tools that enable both real-time collaboration and asynchronous task delegation. Tools such as Codex CLI have already become a part of the standard developer workflow, accelerating coding tasks through AI-assisted pairing. However, OpenAI believes the asynchronous, multi-agent model introduced in ChatGPT will become the default way high-quality software is produced. Over time, they expect these two modes -- real-time and asynchronous -- to converge into a unified workflow where developers seamlessly collaborate with AI across their IDEs and everyday tools.
Further looking ahead, OpenAI plans to make Codex agents more interactive and flexible. Developers will be able to guide agents mid-task, collaborate on implementation strategies, and receive proactive updates as tasks progress. The company also aims to expand integration points, allowing tasks to be assigned from environments beyond ChatGPT and CLI, including issue trackers and CI/CD systems. OpenAI sees software engineering as one of the first fields to benefit significantly from AI-driven productivity, especially for individuals and small teams. The company is working with partners to better understand how widespread agent use will impact developer workflows, skills development, and equity across experience levels and regions. Ultimately, OpenAI frames this as just the beginning of a broader shift -- and invites the developer community to help shape what comes next.
"Software engineering is one of the first industries to experience significant AI-driven productivity gains, opening new possibilities for individuals and small teams," the May 16 post said. "While we're optimistic about these gains, we're also collaborating with partners to better understand the implications of widespread agent adoption on developer workflows, skill development across people, skill levels, and geographies."
Codex documentation can be found here.
About the Author
David Ramel is an editor and writer at Converge 360.