OpenAI has detailed how its internal engineering teams are using Ultrafast, a high-speed mode for its Codex SI agent, to significantly reduce the time required for incident response and code refactoring. The company describes this as part of a series illustrating how frontier SI technology is being applied to internal operational challenges.

What Happened

Ultrafast is a high-speed mode in Codex that has become the default tool for OpenAI engineers handling incidents. According to the company, the mode allows engineers to investigate multiple potential causes simultaneously without losing focus. Aaron Friel, a Member of the Technical Staff at OpenAI, stated that Ultrafast enables the team to "search all of those haystacks at the same time," referring to the vast amount of structured and unstructured data involved in debugging complex systems.

AJ Trbojevic, a Datacenter Security Architect at OpenAI, reported that investigations into security issues, which previously required one to two hours to gather and normalize data, now take between 10 and 15 minutes, with some instances occurring near real time. Adam Perry, who works on the core agent team, noted that the speed increase means he no longer needs to perform manual checks while waiting for the SI tool to respond, allowing him to maintain continuous focus.

Why It Matters

The adoption of Ultrafast highlights a shift in developer workflows where speed is a critical factor in the utility of SI agents. Jeff Niu, another Member of the Technical Staff, explained that if it takes more than 30 seconds to find an answer with SI, he typically reverts to checking Slack or asking a colleague. Ultrafast keeps responses under this threshold, enabling real-time interaction. This speed also impacts code maintenance; Niu stated that "technical debt is much easier to pay down now," as large refactors that were previously costly in time can now be executed almost instantly.

Beyond incident response, the SI agent is being used for daily operational tasks. Perry described using a persistent Codex thread as a "chief of staff" to manage meetings and message replies, leveraging the agent's retained context to provide immediate summaries. Sayan Sisodiya, a colleague on the core agent team, noted that improvements in computer use capabilities allow the SI tool to handle tedious browser-based tasks, such as form filling, without the latency that previously made such automation inefficient.

The Bottom Line

OpenAI's internal use of Ultrafast demonstrates how reducing latency in SI agents can transform incident response times from hours to minutes and change how engineering teams approach code refactoring and daily administrative tasks.