Investigations into security issues used to take an hour or two to pull structured and unstructured data together, normalize it, and provide context to make a decision, according to AJ Trbojevic, Member of the Technical Staff and a Datacenter Security Architect at OpenAI. “Now that happens in 10 to 15 minutes,” as Trbojevic stated. “And at times it’s near real time.”
This shift in operational speed exposes a harsh economic reality for modern tech enterprises where engineering downtime directly erodes margins. When critical systems fail, standard incident response workflows demand substantial human capital to coordinate, diagnose, and resolve errors while revenue-generating infrastructure sits idle. The deployment of automated high-speed diagnostic loops fundamentally alters this P&L equation by compressing expensive engineering hours into negligible operational intervals.
Architecture of Speed
The technological mechanism driving this compression is Ultrafast, a high-speed mode in Codex that has become the default for incident response among engineers at OpenAI, as documented in internal company materials. When infrastructure breaks, waiting minutes for diagnostic output compounds the financial loss through extended downtime. Ultrafast allows engineers to investigate multiple potential causes at once and get answers without splitting their attention.
Aaron Friel, Member of the Technical Staff at OpenAI, who is usually the one running point on an incident, describes the diagnostic bottleneck as navigating a massive data maze. He spins up a Slack channel for incident response and then points Codex at the problem, asking Codex to monitor, investigate, and answer any questions as they come in. “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,” as Friel commented. “Using Ultrafast, I’m able to essentially search all of those haystacks at the same time, and respond to incidents much faster.”
Using Ultrafast, I’m able to essentially search all of those haystacks at the same time, and respond to incidents much faster, as Aaron Friel, Member of the Technical Staff at OpenAI, commented.
Adam Perry and Sayan Sisodiya, both working on the core agent team, build the code that lets Codex take actions, run commands, and use tools. Adam Perry notes that prior to this architecture, he would ask Codex to look into something and then investigate manually at the same time because waiting for the system was not fast enough. With the new speed mode, Codex operates quickly enough to maintain singular focus.
The Financial Impact on Technical Debt
The reduction in latency changes how engineering teams handle long-term code maintenance and operational overhead. Jeff Niu, another Member of the Technical Staff, states that if it takes more than 30 seconds to find an answer with AI, he will check Slack or chat with a coworker, but most of his work is now under that threshold with Ultrafast. Previously, waiting for query results caused context switching that degraded productivity. Now, real-time responses keep engineers locked onto the primary task.
This low-latency interaction model lowers the barrier to addressing technical debt. Jeff Niu notes that Ultrafast makes him more willing to tackle large refactors because reworking existing code takes less time. “Technical debt is much easier to pay down now than it ever was before,” as Niu stated. “Code is cheap, and now it’s very quick to produce. A refactor that used to cost a cost a lot happens almost instantly.”
Does accelerating internal engineering tasks through proprietary APIs create a scalable template for enterprise efficiency, or does it merely highlight a capability gap that commercial customers cannot yet bridge with standard tools?
Automated efficiency extends beyond debugging and refactoring into routine operational tasks that previously consumed valuable engineering bandwidth. Adam Perry keeps a running Codex thread that he calls his chief of staff, which every morning tells him what meetings he has and which messages he might have forgotten to reply to. Sayan Sisodiya uses Codex to do tedious manual work in the browser, like filling out forms—something he’d previously avoided because it was too slow. With improvements in computer use, he is comfortable letting the agent handle interface execution without losing time.
How will software pricing models adapt when internal developer productivity increases by orders of magnitude through autonomous tooling?
Editorial analysis
The leading developer of frontier language models has overhauled how it handles internal infrastructure threats. OpenAI deployed the Ultrafast mode in Codex across its engineering workflows; according to the company, investigating incidents now takes 10 to 15 minutes instead of up to two hours, occasionally running almost in real time. Speed became critical after mid-September, when we reported that researchers compromised the lab's internal accounts in under 72 hours using Anthropic's Claude for just $3,000. Automated triage shrinks response windows, but the race between autonomous offensive tools and defenses is just beginning.
We covered this: August 27, September 18