Ultrafast is a high-speed mode in Codex that has become the default tool for incident response among engineers at OpenAI. The mode is designed so engineers don’t have to wait minutes for answers while a problem escalates; with Ultrafast they can investigate multiple possible causes simultaneously and receive near-real-time responses.
How teams deploy it
Aaron Friel, Member of the Technical Staff, typically leads incident responses: he opens a Slack channel and directs Codex to monitor, investigate, and answer incoming questions. "In any incident there are usually many hops along the way of what could have gone wrong, and finding the right needle in the right haystack is really the chief job of the people on that channel," he says. Using Ultrafast, he can effectively search all those haystacks at once and respond more quickly.
AJ Trbojevic, Member of the Technical Staff and a Datacenter Security Architect, says security investigations used to take an hour or two to pull structured and unstructured data together, normalize it, and provide context. "Now that happens in 10 to 15 minutes," he says, adding that sometimes the response is near real time.
Adam Perry and Sayan Sisodiya, who work on the Core Agent team building the code that lets Codex take actions, run commands, and use tools, have also changed how they operate during security incidents. Adam explains that previously he would ask Codex to look into an issue and simultaneously start his own investigation because Codex wasn’t fast enough to wait on. With Ultrafast, Codex is fast enough that he can stay focused instead of splitting his attention.
Sayan highlights that Ultrafast lets him get immediate answers from multiple data sources: beyond digging through logs, he uses Codex to search Slack for references to suspicious changes or similar symptoms mentioned elsewhere.
Shorter waits, uninterrupted focus
Jeff Niu, Member of the Technical Staff, says that if finding an answer with AI took more than 30 seconds he would check Slack or talk to a colleague. With Ultrafast, most of his work falls below that threshold. He used to ask Codex a question, switch to another task while it thought, and return having lost his train of thought; the near-real-time interaction preserves continuity.
Makes refactors and routine work easier to do
Ultrafast’s speed also makes Jeff more willing to undertake large refactors because reworking existing code takes less time. "Technical debt is much easier to pay down now than it ever was before," he says. Tasks that once required significant time and resources can now happen almost instantly.
Adam maintains a running Codex thread he calls his chief of staff; each morning it summarizes his meetings and points out messages he might have missed. Because the thread has already scanned the data to produce that summary, follow-up questions don’t require fetching more information, allowing a continuous conversation.
Sayan uses Codex to perform tedious manual browser work such as filling out forms—tasks he previously avoided because the tool would pause for several seconds between steps. Improvements in the models’ ability to control the computer have made him comfortable delegating those interactions: Codex clicks through the browser quickly enough that he doesn't lose time by not doing it himself.
Impact on daily work
Across incident response, daily planning, and refactoring, Ultrafast lets OpenAI’s Technical Staff work at the pace of their questions. They can follow up on answers, try alternative approaches, and keep working without losing momentum. The reduced waiting time also changes what they are willing to take on—from previously postponed refactors to routine tasks that now make sense to delegate to Codex.
