OpenAI Introduces Codex: An AI Coding Agent for Developers

May 20, 2025 Ai Binger News Desk

OpenAI has launched Codex, a new AI-powered coding assistant designed to help developers write, manage, and improve code more efficiently.

It is currently available as a research preview for select ChatGPT Pro, Team, and Enterprise users. Access for Plus and Edu users is coming soon.

Key Features

Built into ChatGPT: Codex is accessed through a side panel in ChatGPT. Developers can use simple prompts to generate code, debug, or ask questions about their codebase.

Powered by codex-1: Codex runs on a version of OpenAI’s o3 model tailored for software engineering. It can perform multiple tasks at once—writing features, fixing bugs, and submitting pull requests.

Runs in the cloud: Each task is executed in an isolated cloud environment that mirrors the developer’s project files, ensuring secure and context-aware operations.

Task duration and audit trail: Tasks can take from under a minute to up to 30 minutes. After completion, Codex provides a detailed audit trail showing its actions, terminal output, and test results.

Custom guidance: Developers can include an AGENTS.md file in their repositories to guide Codex with project structure, coding practices, and test commands.

Security-first design: Codex is trained to reject malicious coding requests while supporting legitimate software development.

Current Limitations

No support for image-based tasks
Tasks cannot be paused or interrupted once started
In some cases, Codex may be slower than manual coding
OpenAI plans to add mid-task guidance, deeper integration with developer tools and introduce rate limits and pay-as-you-go pricing.

Background and Adoption

Codex arrives at a time of rising demand for AI-powered development tools, also known as “vibe coders.” Competitors like Google and Anthropic are also investing heavily in this space.

OpenAI recently acquired Windsurf, a major player in AI coding, for $3 billion—signaling how high the stakes are.

Major companies have already adopted Codex:

Cisco: Testing it to accelerate engineering workflows

Superhuman: Using it to improve test coverage and enable code changes by non-engineers

Kodiak: Enhancing code reliability and understanding complex software

Temporal: Automating background tasks like debugging and test writing

Codex is a significant step forward from the original Codex model that powered GitHub Copilot, offering a more integrated and capable solution for software teams.