The Unintended Consequences of Codex’s Comeback

OpenAI’s Codex—which was once pitched as a coding agent for developers and is now being framed as a more general-purpose agent for knowledge workers—is having a moment, as OpenAI chief revenue officer Denise Dresser put it on Tuesday. 

As I wrote about in this story from Tuesday, lots of developers are making the switch from Anthropic’s more popular Claude Code tool to Codex, as OpenAI has made its models and the Codex app better for longer-running and more complex tasks in recent months, they say.

Engineers at OpenAI have seen that productivity boost perhaps more than anyone else. However, that has created its own issues. OpenAI employees are now making so many code changes that it has caused outages in the systems that manage the company’s vast codebase, according to two people with knowledge of the situation.

Many of OpenAI’s thousands of engineers have gone from making two or three code changes a day to more than 10, the people said. Each of those changes trigger thousands of hours worth of tests that run in parallel on many computers, which ensure the code runs correctly and doesn’t have any bugs or security vulnerabilities, they said.