It’s not only tech companies that are saying that large amounts of their code is now being written by AI — their employees are saying it too.
OpenAI employee Aidan McLaughlin has said that 80% of his code is now written by AI. “Fwiw 80% of my code is now written by Codex,” he posted on X. “And I’m writing a lot more code,” he added.
Codex is OpenAI’s recently-released software engineering agent. It is based in the cloud, and can work on many tasks in parallel. Codex is powered by Codex-1, a model that OpenAI has optimized for coding tasks.
McLaughlin’s assertion would make it appear that AI can now write fairly advanced code that’s being used at the top AI labs — he works at OpenAI on model behavior and capabilities research. But he isn’t the only top AI lab employee that says that a large fraction of their code is being written by AI. An Anthropic employee has said that 80% of the code for Claude Code, Anthropic’s coding agent, was being written by Claude Code itself. Bigger companies have given slightly more conservative estimates of how much of their coding has been automated — Google says that well over 30% of its code is now being written by AI, while Microsoft says that 20-30% of its code is AI generated.
And this is the path to superintelligence that many have predicted. If code at the top AI labs is being written largely by AI, it can increase the research output of these labs to a great degree — McLaughlin, for instance, says he’s writing more code than before. This would help quickly build and improve the AI that’s generating the code, which in turn would be able to write code even faster. This could create a flywheel that will dramatically speed up AI research, and lead to new discoveries and paradigms. And with top AI companies and their employees already talking about how much AI is helping them with their jobs, superintelligence might be closer than we think.