AIToday
Large Language ModelsAI Coding AssistantsGIGAZINE AIPublished: Nov 20, 2026, 19:01 JST

OpenAI's Codex-powered Performance Factory runs 1,700 investigations in 3 months

Send on LINE
OpenAI's Codex-powered Performance Factory runs 1,700 investigations in 3 months

3 Key Points

  1. What happened

    OpenAI's Performance Factory, built on Codex, completed over 1,700 investigations in three months and pinpointed root causes in more than 600 cases, coordinating agents across four components including task service and dispatch loop.

  2. Why it matters

    The setup lets agents assign tasks, run tests, and deploy to production autonomously, shifting engineers from hands-on troubleshooting to reviewing fixes and could reduce the manual burden of maintaining service speed.

WHO IT HITSEngineering and site-reliability teams that maintain high-traffic online services stand to benefit, as the Performance Factory shows how agents can take over routine investigation and fix-proposal work that once required manual effort.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

OpenAI's post traces a clear progression in its internal engineering workflow. Early on, human engineers handled problem investigation and countermeasures by hand; Codex then became a partner, but humans still led the work. As AI capability and skills matured, Codex sessions began starting automatically when alerts fired and posting findings to Slack. The current stage is an "agent factory" that autonomously identifies needed tasks, assigns agents, runs tests, and deploys to production. The Performance Factory, built on four components—task service, dispatch loop, cloud Codex, and subscription—is the concrete implementation of that stage for keeping ChatGPT fast. The reported results—over 1,700 investigations and more than 600 root-cause identifications in three months—suggest the system is already handling a substantial volume of diagnostic work. OpenAI says it is now working to make the system proactive, finding slowdowns and optimization opportunities even without alerts, which would extend its role from reactive troubleshooting to continuous performance tuning.

FAQ
What are the four components of the Performance Factory?
They are a task service that structures tasks, a dispatch loop that assigns ready tasks to agents, cloud Codex that executes tasks, and a subscription that triggers agents when a change occurs.
When does the Performance Factory currently operate?
It runs when alerts about slowdowns occur; OpenAI is working to make it proactively find performance drops and optimization areas even without alerts.
How has OpenAI's development process evolved with Codex?
It started with human engineers doing manual investigation, then Codex as a partner with humans leading, then autonomous Codex sessions that post findings to Slack, and now an agent factory handling task identification, assignment, testing, and deployment.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleDayOne shelves $5 billion US IPO on AI valuation jitters