AI Agents for Performance: Ship Faster, Pay Less
summary
An agent reads a profiler call stack, traces a hot path to the exact source line, and turns an quadratic-time tensor merge into an O(N) fix that is validated by canary validation on real production traffic with zero regression allowed before shipping. The system checks out the running commit, runs tests, opens a code review for human approval, and surfaces the same antipattern across seven services into a shared pattern catalog for future coding agents.
AI-written summary · about
chapters (15)
- 0:00Introduction: performance for ML serving
- 2:24The manual profiling loop today
- 4:46The experiment: can an LLM read a profile?
- 7:37From call stack to the exact method
- 9:07How the agent locates and reads the code
- 11:01First finding: an O(N) fix, canary confirmed
- 12:41The same antipattern across seven services
- 15:31Building a shared pattern catalog
- 17:16Storing and sharing findings across services
- 19:40Feeding the catalog to coding agents
- 20:56Human approval and verification
- 22:13Canary validation on real traffic
- 24:18Reactive vs proactive paths
- 26:54Catching waste before production
- 29:32Autonomy levels and what is next
related talks
21mRemote Agents, or: How I Fixed Our Page's Performance Without Really Doing Anything
27mThe AI Velocity Trap: Shipping Faster Without Breaking More
30mBeyond the Hype: Using AI Coding Agents Today
26mAI Agents as Your Co-Developer
45mAI agents in practice - beyond the prototype
16mAI Agents for Incident Investigation: The Good, The Bad, and The Ugly
07h 47mAI Agents Conference 2026 - Day 5
26mFrom autocomplete to agents: AI coding assistance state of play