Ultrafast, a high-speed mode in Codex, has become the default tool for incident response at OpenAI. Engineers no longer need to wait minutes for answers while issues compound. Instead, they can investigate multiple potential causes simultaneously without splitting their attention.
Aaron Friel, Member of the Technical Staff at OpenAI, leads incident response by spinning up Slack channels and pointing Codex at the problem to monitor, investigate, and answer questions in real time. "Using Ultrafast, I'm able to essentially search all of those haystacks at the same time," he says.
AJ Trbojevic, Datacenter Security Architect, notes that investigations once taking an hour or two now happen in 10 to 15 minutes—sometimes near real time. Adam Perry and Sayan Sisodiya on the core agent team report that Ultrafast allows them to stay focused rather than duplicating efforts.
Jeff Niu, another Member of the Technical Staff, says responses under 30 seconds keep him engaged. Ultrafast also makes large refactors feasible, with technical debt easier to pay down. Codex acts as a copilot, summarizing meetings and messages each morning.
Source: OpenAI Blog · Summarized by HeadlinesBriefing