Almost All Code at OpenAI Is Now Written by AI
At OpenAI’s DevDay 2025, CEO Sam Altman made a statement that captured the tech world’s attention: “Almost all new code written at OpenAI today is written by Codex users.” This announcement signals a major shift in how the world’s most advanced AI company builds its own technology — using AI to write AI.
Codex: The New Developer in the Room
Codex, OpenAI’s AI coding assistant, has quickly become central to its internal engineering process. According to Altman, developers at the company are now completing about 70% more pull requests per week than before. Codex not only generates code but also reviews it, providing consistent feedback across all levels of engineering.
What’s even more remarkable is the speed at which new projects are being developed. The new drag-and-drop “Agent Builder” tool, for instance, was created in under six weeks — with Codex writing roughly 80% of the pull requests. This accelerated pace of innovation is changing how OpenAI manages software development altogether.
The scale of Codex’s usage is staggering. Its latest model, GPT-5-Codex, has processed over 40 trillion tokens within three weeks of launch. Daily activity has grown tenfold since August, proving that AI-assisted coding isn’t just a novelty — it’s rapidly becoming the norm inside OpenAI.
Empowering Developers and Enterprises
To make Codex more accessible, OpenAI introduced several new features during DevDay 2025. Admins can now manage and monitor Codex environments, track performance, and set quality thresholds for generated code. Developers can use the Codex SDK to integrate AI-powered code generation directly into their applications or workflows.
Codex also works within popular tools like GitHub and Slack, making it easier for teams to access AI coding assistance where they already collaborate. The goal is clear: to make AI a seamless part of the software engineering experience, not just an optional add-on.
Changing the Definition of Programming
The implications of OpenAI’s internal adoption are huge. If most of the code at one of the world’s leading AI labs is now written by an AI system, it challenges traditional notions of what programming even means. Developers are shifting from manually writing every line of code to managing, refining, and validating AI-generated work.
However, the transition isn’t without concerns. Surveys among developers show that while the majority now use AI tools, fewer trust the accuracy or maintainability of AI-generated code compared to last year. Questions of security, data privacy, and long-term code quality remain open.
The Future of AI-Driven Development
Despite these challenges, companies across industries — from e-commerce and education to cybersecurity — are beginning to follow OpenAI’s lead. Codex is helping them automate repetitive coding tasks, speed up code reviews, and focus human creativity on solving complex problems.
Altman’s declaration at DevDay 2025 underscores a profound truth: the way we build software is evolving faster than ever. AI isn’t just writing code — it’s reshaping the entire culture of creation. In this new era, the best developers may not be those who write the most code, but those who know how to collaborate with intelligent machines most effectively.
