How OpenAI is using GPT-5 Codex to improve the AI tool itself

In a recent discussion with Ars Technica, OpenAI team members unveiled how integral their AI coding assistant, Codex, has become in enhancing its own capabilities. Alexander Embiricos, the product lead for Codex, shared insights on Tuesday, stating, "The majority of Codex is essentially developed by Codex itself, making it largely self-sustaining in its improvements." Launched as a research preview in May 2025, Codex is a cloud-based software engineering tool designed to tackle an array of programming tasks, including feature development, bug fixing, and pull request proposals. It operates in secure environments that are directly connected to a user’s code repository, enabling it to perform multiple tasks concurrently. Users can access Codex via various platforms, such as the web interface of ChatGPT, command-line interface (CLI), and IDE extensions compatible with Visual Studio Code, Cursor, and Windsurf. The name “Codex” traces back to an earlier OpenAI model from 2021, which powered GitHub Copilot’s tab completion feature. According to Embiricos, there is speculation among staff that “Codex” is an abbreviation for “code execution.” This branding aims to link the current version to its predecessor, which was pivotal in showcasing AI's potential to understand user context and enhance productivity. Interestingly, the existing command-line version of Codex shares similarities with Claude Code, Anthropic’s coding assistant launched in February 2025. When questioned about the influence of Claude Code on Codex's design, Embiricos refrained from confirming any direct impact but acknowledged the vibrant competition in the market. "It’s an exciting space to be in, filled with innovative ideas," he remarked, adding that OpenAI had been internally developing web-based features for Codex before releasing the CLI version, which came after Anthropic’s tool.