Ultrafast如何将OpenAI事件调查从几个小时减少到几分钟
这是我们系列的一部分，分享OpenAI如何使用自己的技术和API的内部实例。 这些工具正在OpenAI内部使用，并在此共享，以作为示范实例，说明边境AI如何支持我们各团队的使用情况。
Ultrafast是Codex中的一种高速模式，对于OpenAI的工程师来说，当事情破裂时，工程师们往往没有时间等待回答，利用Ultrafast，他们可以同时调查多种潜在原因，并获得答案而不会分散他们的注意力。
OpenAI技术工作人员成员Aaron Friel通常是事件的一个运行点。 他旋转了Slack频道进行事件响应，然后将Codex指向问题，要求Codex在进入时监视，调查和回答任何问题。 “无论如何，在可能出错的道路上通常有许多跳跃，在适当的干草堆中找到正确的针头，确实是该频道上人们的首要工作。” 我能够同时搜索所有这些干草堆
AJ Trbojevic，技术工作人员和Datacenter安全建筑师，称对安全问题的调查，通常需要一到两个小时才能将结构化和无结构化的数据拉在一起，使其正常化，并提供作出决定的背景。 “现在这发生在10至15分钟内，” “有时它几乎是实时的。”
亚当·佩里和萨扬·西索迪亚都为核心的agent团队工作，构建了让Codex采取行动，运行命令，以及使用工具的代码。 Ultrafast公司在安全事件中改变了运作方式。 「在此之前，我要求Codex调查一些事情，然后自己去调查，因为我知道它不够快，“现在随着Ultrafast,Codex已经足够快了，我只能集中精神，而不是分散注意力。”
Sayan欣赏Ultrafast如何让他从多个数据来源获得即时答案：“在一次事件中
如果需要30多秒时间才能找到答案，但是他的大部分作品现在都和"超快"一起处于这一门槛之下。 他经常问克德克思一个问题，在它思考的时候转机给别的东西，在失去他的思想火车后回来。 现在这种对话是实时发生的；这帮助他专注于手头的任务。
Ultrafast也使杰夫更愿意去处理大的重构物，因为对现有代码的再处理需要的时间更少。 “技术债务现在比以前容易偿还。” “守则很便宜，一个曾经花费很多钱的重构几乎立即发生。”
亚当保留了一条运行 Codex 线，他称他的参谋长。 每天早上，它都告诉他有什么会议，以及他可能忘记回答哪些信息。 称表示：「它已经查看了所有数据来撰写摘要，“我只能像真正的参谋长一样，在同一次谈话中继续提问。”
Sayan使用Codex在浏览器中做乏味的手工活，“它会打开浏览器，坐在那里想五秒钟，点击一些东西，再坐在那里，” 随着最新模型在计算机使用方面的改进，他更愿意让Codex处理这项工作。 “现在我完全可以让它点击浏览器
Ultrafast让OpenAI的技术人员按照问题的速度与Codex合作。 他们可以追踪一个答案，尝试另一种方法，继续工作而不失去他们的思维。 如此短的等期也改变了他们愿意承担的
- 密码
- 2026
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
How Ultrafast cut incident investigation from hours to minutes at OpenAI
This is part of our series sharing internal examples of how OpenAI is using its own technology and APIs. These tools are being used internally, at OpenAI, and are shared here as illustrative examples of how frontier AI is supporting use cases across our teams.
Ultrafast is a high-speed mode in Codex, and for engineers at OpenAI, it’s become the default for incident response. When something breaks, engineers often don’t have time to wait minutes for an answer while the issue compounds. With Ultrafast, they can investigate multiple potential causes at once and get answers without splitting their attention.
Aaron Friel, Member of the Technical Staff at OpenAI, is usually the one running point on an incident. 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,” he says. “Using Ultrafast, I’m able to essentially search all of those haystacks at the same time, and respond to incidents much faster.”
AJ Trbojevic, Member of the Technical Staff and a Datacenter Security Architect, says 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. “Now that happens in 10 to 15 minutes,” he says. “And at times it’s near real time.”
Adam Perry and Sayan Sisodiya both work on the core agent team, building the code that lets Codex take actions, run commands, and use tools. Ultrafast has changed how they operate during security incidents. “Before, I’d ask Codex to look into something, and then go look into it myself at the same time, because I knew it wouldn’t be fast enough to just wait on,” Adam says. “Now with Ultrafast, Codex is fast enough that I can just stay focused instead of splitting my attention.”
Sayan appreciates how Ultrafast lets him get immediate answers from multiple data sources: “Beyond just digging through logs in an incident, I use Codex to dig through Slack to see if there are references to changes that might be suspect or people talking about similar symptoms elsewhere in Slack,” he says.
If it takes more than 30 seconds to find an answer with AI, Jeff Niu, another Member of the Technical Staff, says he’ll check Slack or chat with a coworker. But most of his work is now under that threshold with Ultrafast. He used to ask Codex a question, switch over to something else while it thought, and come back having lost his train of thought. Now that conversation happens in real time; it helps him stay focused on the task at hand.
Ultrafast also makes Jeff 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,” he says. “Code is cheap, and now it’s very quick to produce. A refactor that used to cost a lot happens almost instantly.”
Adam keeps a running Codex thread that he calls his chief of staff. Every morning it tells him what meetings he has and which messages he might have forgotten to reply to. “It’s already looked at all the data to write that summary, so when I ask a followup, it’s not waiting to go fetch more,” he says. “I can just keep asking questions in the same conversation just like an actual chief of staff.”
Sayan uses Codex to do tedious manual work in the browser, like filling out forms—something he’d previously avoided because it was too slow. “It would open a browser, sit there thinking for five seconds, click something, sit there again,” he says. With the latest models’ improvements in computer use, he’s more comfortable letting Codex handle that work. “Now I’m totally fine letting it click through the browser and it’s fast enough that I’m not losing time by not doing it myself.”
Across incident response, daily planning, and refactoring, Ultrafast lets OpenAI’s Technical Staff work with Codex at the pace of their questions. They can follow up on an answer, try another approach, and keep working without losing their train of thought. That shorter wait also changes what they’re willing to take on, from a refactor they’d put off to a routine task that’s finally worth delegating.
- Codex
- 2026
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.