OpenAI Runs Codex Agent Factory to Keep ChatGPT Fast
OpenAI published an explanation of how it uses Codex to maintain ChatGPT's speed.
1 source 50m ago
Reporting from 1 source: GIGAZINE.
Why it matters
OpenAI now lets Codex agents handle task assignment, testing, and production deployment for ChatGPT performance work, and says it wants the system to find slowdowns before alerts fire.
Key Facts
- OpenAI published an explanation of how it uses Codex to maintain ChatGPT's speed.
- The internal system is called the Performance Factory and has four components: Task Service, Dispatch Loop, Cloud Codex, and Subscriptions.
- OpenAI says the Performance Factory ran more than 1,700 investigations in three months.
- OpenAI says the system identified root causes in more than 600 cases.
- The Performance Factory currently operates when alerts about ChatGPT speed degradation occur.
The internal system, called the Performance Factory, has four parts: a Task Service that composes tasks, a Dispatch Loop that assigns ready tasks to agents, Cloud Codex that executes them, and Subscriptions that launch agents when a change occurs. OpenAI says the factory ran more than 1,700 investigations in three months and found root causes in more than 600 cases.
The Performance Factory grew out of a manual process. Engineers first investigated problems and wrote fixes by hand. Codex later worked alongside them, but humans still led the workflow. As the model improved, Codex took over more of the work, and a session now starts automatically when an alert fires and posts its findings to Slack.
At the time of writing, the factory runs when alerts about ChatGPT speed degradation appear. OpenAI says it is working to make the system find performance degradation and optimization areas even without an alert.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google