OpenAI Performance Factory automates ChatGPT regression fixes
Original title: How OpenAI’s Performance Factory keeps ChatGPT fast
In just over three months the Regression Workflow tracked more than 1,700 investigations and root-caused over 600.
OpenAI described Performance Factory, a cloud system that uses Codex agents to investigate and fix ChatGPT performance regressions. An alert opens a task; an agent diagnoses the slowdown and proposes a fix for an engineer to review, then a later agent checks whether performance returned to baseline. The first workflow, Regression Workflow, covers that path from alert to verification.
- OpenAI reports 17 automated investigations a day.
- Investigation steps were captured in a shared skill called perf-forensics.
- The system has a task service, dispatch loop, cloud Codex, and subscriptions.
- Proposed fixes go to an engineer for review.
From the original article
14 min read at openai.comLearn how OpenAI built the Performance Factory, a system of Codex-powered workflows that finds and fixes performance issues to help keep ChatGPT fast.
openai.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.