OpenAI wants you to see its own coding tool at work on its own code.
The company published a guide describing how its engineering teams actually use Codex day to day: understanding unfamiliar parts of the codebase, refactoring systems that have accumulated cruft, tracking down performance problems, and generally moving faster through routine engineering work. It's not a product launch - there's no new model version or pricing change attached. Instead it reads as an internal case study, posted to OpenAI's business resources page, aimed at other companies sizing up Codex for their own teams.
The real value here isn't the headline claim that OpenAI's engineers use Codex - of course they do. It's the specific jobs OpenAI is willing to put its name on: comprehension and refactoring, the unglamorous maintenance work that eats far more engineering time than writing new features ever does.
Treat this the way you'd treat any vendor's dogfooding story. GitHub has made similar claims about Copilot inside Microsoft, and Google has talked up Gemini use in its own codebase - internal adoption anecdotes are marketing collateral dressed as documentation, and this one arrives with no numbers attached to check against.