Generic AI prompts burn tokens guessing at performance issues from raw source code 📉 Chrome DevTools for agents fixes this. Let your coding agent run real performance traces and agentic Lighthouse audits to catch actual speed, accessibility, and SEO blockers directly in the browser → #GoogleIO

real perf traces beat guessing from raw source. the gap left is the agent driving your actual logged-in browser, not a headless instance blind to your session. we built fazm to drive the real browser via an extension for exactly that. written with ai

This is the right direction for coding agents: less guessing from static code, more grounded feedback from traces, audits, and runtime signals.

95th percentile matters a lot more than median when you're actually shipping.

So I am new to the vibe coding how to use Chrome dev tools with antigravity or Claude 🥲

This is a fake browser extension impersonating TronLink. The fake extension steals users' private keys, seed phrases (mnemonics), and engages in other malicious behavior. Please take it down / remove it from the store as soon as possible.

Absolute life and time saver especially for annoying bugs that are hard to catch just by looking at the code. With chrome dev tools you maximize the power of the inspection tool in an automated secure sandboxed environment. Who ever made this I owe you a beer.

This is the right kind of agent context: not more prompt poetry, actual traces. Performance bugs do not care how confident the model sounds.

WTF🚨‼️@Google is literally copying our github repo,

Tracing and Lighthouse in a real browser beats guessing Interaction to Next Paint from a diff alone. That fits the coding loop that DevTools for agents is aimed at. After a merge, we still schedule Lighthouse on production URLs and alert when Core Web Vitals budgets break. One local trace does not tell you checkout drifted on a client site you checked weeks ago. We wrote up when one-off PageSpeed Insights stops scaling for multi-site teams: