Introducing Codex: OpenAI’s Advanced AI Coding Assistant

Codex, the latest AI coding assistant from OpenAI, has been released as a research preview for subscribers of ChatGPT Pro, Enterprise, and Team packages. This powerful tool is powered by the most recent o3 AI reasoning model, meticulously fine-tuned for software engineering. Codex is tailored to function as a virtual coworker, supporting developers in writing code, debugging, running tests, and managing various software engineering tasks concurrently, all within a cloud environment.

How Codex Works in Software Engineering

Codex operates in a secured virtual environment that is preloaded with users’ code repositories through GitHub integration. This ensures that it accurately mirrors individual development environments. Users can engage with Codex via the ChatGPT sidebar by providing prompts, allowing them either to generate code or inquire about specific aspects of their codebase. The AI executes its tasks in isolated containers, with execution durations that vary from one to thirty minutes based on the complexity of the task.

Features of Codex

A standout characteristic of Codex is its transparent workflow. It showcases its thought process step-by-step, offers citations of terminal logs as proof of its actions, and communicates any uncertainties or failures during tests. This transparency is designed to cultivate trust and enhance the reliability of code, effectively addressing common criticisms of previous AI coding tools that occasionally produced unclear or non-compliant scripts.

Customising Codex for Your Needs

Developers have the option to tailor Codex’s behaviour by adding an “AGENTS.md” file to their repositories. This file provides the AI with specific directives regarding code context, style guides, and best practices, functioning like a README for AI agents.

Insights from OpenAI’s Leadership

Sam Altman, the CEO of OpenAI, highlighted the transformative potential of Codex on social media, underlining that it can perform multiple tasks simultaneously. This capability significantly accelerates engineering workflows. Codex is equipped to read and edit files and can execute commands such as test harnesses, linters, and type checkers.

“Today we are introducing Codex. It is a software engineering agent that runs in the cloud and undertakes tasks for you, such as writing new features or fixing bugs. You can execute numerous tasks in parallel.” — Sam Altman (@sama) May 16, 2025

Availability and Future Plans

At this time, Codex is available to eligible ChatGPT users, with generous free usage to explore its features. However, OpenAI has plans to implement rate limits and introduce a credit-based pricing model in the upcoming weeks. The company also aims to expand access to ChatGPT Plus and educational users in the near future.

Importance of Human Oversight

While Codex automates a wide range of programming tasks, OpenAI emphasises that human developers must conduct manual reviews and validations to ensure the quality and security of the code.