Netflix used an AI agent to trace a CPU waste bug to one line and validate the fix in canary
AI Engineer · youtube · 2026-07-28
Netflix engineer Rajat Shah describes how the team used an AI agent to find and fix a hidden performance bug in production.
The agent takes a call stack, traces the issue to the exact source line, proposes a fix, and validates it against a canary traffic split before shipping. In the case highlighted in the talk, it discovered an inefficient quadratic-time pattern in a live service that had been quietly burning CPU on every request.
Key takeaways:
- Traditional profiling loops are slow, so waste can sit in production for months.
- AI-assisted coding can worsen this by producing code that is fast to write, not fast to run.
- The team wants a shared catalog of recurring performance antipatterns so they can be caught earlier.
- The same pattern surfaced in seven other services in one session, suggesting a scalable detection loop.
- The next step is feeding that catalog back into coding agents before code ships.
More from coding & agent
- Custom scripts often beat waiting for a better MCP, says one AI workflow user — DavidWells · 2026-07-28
- Vibecoding a custom desktop app now also means adding AI image gen in one shot — LauraModiano · 2026-07-28
- IBM launches a free 1-hour graph engineering course for AI agents — goyalshaliniuk · 2026-07-28
- Gemini CLI nightly 0.54.0 fixes CRLF handling and keychain tag validation — gemini-cli-robot · 2026-07-28
- Lessons from a 4AM Posting Spree: Architecting a Fully Autonomous Social Media Agent — Suitable-Nerve-9041 · 2026-07-28
- Opinion: Agents dynamically building specialized harnesses on the fly will be huge — burny_tech · 2026-07-28