Ultrafast cuts OpenAI incident probes to 10–15 minutes

OpenAI is presenting this as an internal case study of how its own teams use Codex, not a product announcement. The company frames Ultrafast as fast enough that engineers can stay in a single conversation instead of switching tasks while waiting for an answer.

Beyond incident response, staff describe using the same Codex threads for daily planning and large refactors. Adam Perry keeps a running thread he calls his chief of staff, and Jeff Niu says technical debt is easier to pay down because refactors that once cost a lot happen almost instantly. Sayan Sisodiya uses Codex for browser form-filling, which he had avoided as too slow.

The through-line the body draws is behavioral, not technical: shorter waits appear to change what staff are willing to delegate or tackle, from a put-off refactor to a routine task.