OpenAI's Codex-powered 'Performance Factory' has run 1,700+ regression investigations in 3 months

OpenAIDevs · x · 2026-11-20

OpenAI's developer team details its internal Performance Factory: Codex agents now investigate ChatGPT performance regressions, propose fixes for engineers to review, and verify in production that fixes restore baseline performance. The companion thread reveals the regression workflow has tracked 1,700+ investigations in just over three months, identifying root causes in 600+ cases. Engineers capture the performance team's playbook in a shared skill ("perf-forensics"), investigations run in the cloud so alerts auto-start Codex sessions, and a task service with subscriptions lets work pass between agents and engineers across code reviews and deployments.

Related event: OpenAI's Codex-powered 'Performance Factory' has run 1,700+ regression investigations in 3 months(3 posts)→

Original post →

More from coding & agent

coding & agent channel →