OpenAI says Codex Ultrafast speeds incident response
OpenAI says its engineers now use Codex's high-speed Ultrafast mode by default for incident response, and staff report that investigations once taking hours now finish in minutes.
OpenAI described how its technical staff use Ultrafast, a high-speed mode in Codex. The post is part of the company's series on internal use of its own tools. It is a self-reported case study. It is not a product launch and includes no benchmarks or outside verification.
The main claim concerns incident response. AJ Trbojevic, a datacenter security architect, said security investigations used to take "an hour or two" to gather and normalize structured and unstructured data. "Now that happens in 10 to 15 minutes," he said, "and at times it's near real time."
Aaron Friel, who often leads incidents, opens a Slack channel and has Codex monitor the problem, investigate it and answer questions. He said Ultrafast lets him "search all of those haystacks at the same time." Sayan Sisodiya of the core agent team uses it to search both logs and Slack for suspect changes or similar symptoms.
Several engineers said the main benefit is that they no longer have to split their attention. Adam Perry said he used to investigate in parallel with Codex because it "wouldn't be fast enough to just wait on." Jeff Niu said that if an AI answer takes more than 30 seconds, he asks a colleague instead, and that most of his work now comes in under that limit. He added that faster output makes large refactors cheaper.
Other uses mentioned include a daily "chief of staff" thread and browser form-filling, which Sisodiya credits to better computer use in recent models.
OpenAI did not say how fast Ultrafast is, how it is priced or which models it runs on.