AI Agents for Performance: Ship Faster, Pay Less — Rajat Shah, Netflix

A performance agent needs evidence of where running software spends time before it proposes an optimization. This Netflix practitioner account describes feeding execution profiles to an agent, locating the corresponding code at the exact deployed version, and preparing a proposed fix.

The workflow then uses functional tests and a limited deployment comparison to check CPU use, response time and errors before an engineer approves the change. The speaker also describes finding repeated creation of a metrics counter across seven services. Recording such patterns in a shared catalog lets later investigations reuse the finding, while each proposed change still needs validation in its own service.

“People typically end up looking at profiling data only when something is going wrong at 2:00 a.m. and somebody needs to fix a problem because your CPU is unbearable.”

Rajat Shah

“Profiler gives the estimate, canary gives ground truth.”

Rajat Shah

“You can't simply run AI agents without the right level of security guardrails so that prompt injection and other security attacks that a typical agent infrastructure currently cannot always solve for.”

Rajat Shah

“Your um agent doesn't know specific details about your platforms and your frameworks uh and your internal code base patterns.”

Rajat Shah

“together a stateful catalog and a state stateless LLM can become a full fleet-wide memory”

Rajat Shah