Devtools, AI, Insights
summary
A Performance AI assistant runs on a fork of Chrome DevTools that parses trace files, teaches LLMs specific telemetry semantics, and wires specialized agents to a custom MCP server and Vzero for code generation. It builds automated attribution over known metrics using a trace engine, narrows context windows, and steers sub‑agents to act. A live demo reports INP down from ~512ms to 29ms, showing the system can meaningfully change outcomes when prompts, routing, and focused signals are correct.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 2:52Demo setup and tooling
- 6:14AI in development workflows
- 14:09Performance intelligence deep dive
- 23:27AI limitations with performance data
- 30:40Trace file analysis and agent design
- 38:34MCP for performance insights
- 43:12Responsible AI integration
- 45:28Live demo results
- 46:10Q&A
related talks
30mMCP DevTools: Giving Your AI Eyes
53mAI Didn’t Kill the Web, It Moved in!
28mDevTools & The Performance Tab
23mBuilding Agent Interfaces: Lessons from Chrome DevTools (MCP) for Agents
26mFrom autocomplete to agents: AI coding assistance state of play
30mThe Chrome DevTools Performance Panel
42mDevTools Deep Dive
01h 00mData is out, knowing is in: Applying AI to automate cloud operations