AI Assistance Becomes Agentic

▶ Watch (2:30)

Earlier AI tools worked like chess turns: write a prompt, get an answer, repeat. Agentic coding tools like Gemini CLI act on a single prompt. Chrome DevTools now brings that same power. In the AI assistants panel, a context selection agent picks the right page context. A prompt like “How can I improve the performance of this page?” triggers Gemini to reload the page, record a performance trace, and analyze Core Web Vitals (LCP, INP, CLS). Responses now average 68% fewer words, getting developers to action faster instead of reading walls of text.

Structured Walkthroughs Keep Developers in Control

▶ Watch (4:58)

Developers want to understand Gemini’s actions. The new agent walkthrough opens in a split view inside the AI assistant panel. Key steps appear as specialized widgets showing computed styles, Core Web Vitals, simplified DOM trees, LCP breakdowns, and first- and third-party performance summaries. Clicking “review” jumps to the exact DevTools panel where that data originates. This replicates Gemini’s steps for human investigation. Six widgets were added for common debugging journeys, with more to come.

Expanded Scope: Lighthouse Accessibility and Export for Agents

▶ Watch (6:41)

AI assistance now accesses Lighthouse scores, enabling new categories of questions. Asking “How can I improve the accessibility of this page?” runs a Lighthouse audit. Gemini provides actionable advice, and can apply quick fixes like color contrast adjustments. After verifying the fix, the “copy for your coding agent” button condenses all investigation results and potential fixes into a ready-to-use prompt. That prompt can be pasted into Gemini CLI or any other coding agent to apply changes to the actual codebase.

WebMCP Debugging: Supporting the Agentic Web

▶ Watch (8:54)

WebMCP is a proposed web standard that lets websites give AI agents specialized tools. DevTools now includes WebMCP debugging tools in the applications panel. Developers can inspect each tool, track invocations, and invoke tools manually to debug results without an agent. This follows DevTools’ pattern of supporting newest web platform features—from view transitions to CSS carousels—before they become baseline.

Notable Quotes

We now bring the same power of agentic conversations to AI assistants in Chrome DevTools. Matthias Rohmer · ▶ Watch (2:22)

On average, we reduced the number of words generated by Gemini to answer your prompt by 68%. Matthias Rohmer · ▶ Watch (4:07)

This makes it easy to follow Gemini’s steps in DevTools and replicate them yourself, too. Matthias Rohmer · ▶ Watch (5:30)

WebMCP is a proposed web standard which enables websites to give AI agents visiting them specialized tools. Matthias Rohmer · ▶ Watch (8:54)

Key Takeaways

  • Agentic AI in DevTools runs performance traces and Lighthouse audits with one prompt, cutting response length by 68%.
  • Structured walkthroughs with widgets let developers inspect Gemini’s actions and replicate steps.
  • WebMCP debugging tools added to support the agentic web standard.