Submind YouTube summaries
Thumbnail for Lighthouse audits with DevTools for agents

Lighthouse audits with DevTools for agents

Watch on YouTube

Video summary

The video introduces an updated version of the open-source tool Lighthouse, now integrated directly into Chrome DevTools for Agents to enhance web development workflows. While traditional maritime lighthouses have evolved over centuries, this digital utility remains a critical asset for developers by providing automated audits across five key categories: performance, accessibility, SEO, best practices, and a new experimental browsing category designed specifically for AI agents. This integration allows coding agents to operate autonomously with deep context from over 140 checks, eliminating the need for manual data transfer between DevTools panels and agent environments. In terms of functionality, the tool offers specialized insights that go beyond simple scoring systems; for instance, performance audits provide raw runtime tracing data rather than just a single score, giving agents precise details to optimize speed. The system streamlines quality assurance by enabling developers to prompt their coding agents to run specific checks, such as accessibility or SEO evaluations, and immediately fix identified issues like missing meta descriptions, poor color contrast, or broken heading structures without human intervention. This closed feedback loop ensures that problems are detected and resolved in real-time before code is pushed to production, significantly reducing the time spent on manual debugging and verification. A particularly significant addition highlighted in the video is the new agentic browsing audit category, which addresses unique challenges faced by AI agents navigating websites. Agents rely heavily on stable layouts, well-formed accessibility trees, and descriptive tools to interact with web content; any unexpected layout shifts or malformed elements can break an entire workflow. By running these specific audits, developers can ensure their sites are robust enough for the emerging agentic web, where a coding agent might identify issues such as missing tool definitions in single-page applications or inaccessible car configurators before they impact automated tasks. Ultimately, Lighthouse serves as a powerful bridge between human oversight and autonomous AI capabilities, allowing teams to meet tight deadlines while maintaining high standards of quality and inclusivity. The video concludes by encouraging developers to explore these new use cases through official documentation and share their own experiences with coding agents surfacing issues via Lighthouse audits. Whether optimizing for search engine crawlers or ensuring seamless navigation for automated bots, this tool provides the necessary context and actionable advice to build faster, more reliable web applications that work seamlessly for all users and machines alike.
Read the full video transcript
Did you know the first lighthouse in recorded history, the Pharos of Alexandria, was built in the 3rd century BCE? Over 23 centuries later, maritime lighthouses might feel a bit outdated for actual nautical navigation. But the lighthouse tool that developers rely on, that is still fresh and is now available in DevTools for Agents, including the new experimental browsing audit category, all built by the same team that maintains Chrome DevTools. Welcome to Developer Tooling Tips. I'm Mathias, let's dive in. >> [music] [music] >> As a quick recap, Lighthouse is an open-source tool helping you improve the quality of web pages across a set of five distinct categories: performance, accessibility, SEO, best practices, and now new experimental browsing. You can run Lighthouse through the DevTools UI in the dedicated panel, through the command line, or as we'll explore today, through Chrome DevTools for Agents. Across these categories, Lighthouse packs a massive punch with over 140 automated checks total, with a straightforward advice. Excellent starting ground as context for your coding agent. For performance, DevTools for Agents has specialized performance tracing tools instead of Lighthouse, giving your agent the raw, more accurate runtime data than the single 0 to 100 score familiar from Lighthouse. And performance insights as part of the trace bring the same context as the original Lighthouse audits. Learn more about them through this link. By using Lighthouse, your agent gets targeted advice based on the live runtime completely autonomously. And after it's done, use the info from those audits to immediately start working on failed audits instead of waiting on you to manually copy-paste them from DevTools into your agent. Let's explore some examples. When you're working on a page intensively, trying to meet deadlines and fulfill client requirements, accessibility is sometimes, even unintentionally, deprioritized. But, we want to make sure your website works seamlessly for as many people as possible. So, after finalizing a change, you can ask your agent to do a final accessibility check with Lighthouse and fix up your work before you push it to your users. In your coding agent, just prompt, "Please run a Lighthouse accessibility audit on this URL." In this example, with our demo news site, DevTools for agents identifies missing button names, bad color contrast, images without alternative text, and non-ideal touch targets. With a follow-up prompt like, "Fix all findings," we can send our coding agent straight to work and have it fix those findings on its own without any manual intervention other than the final code review. On to the next use case. When building a web app, it is easy to overlook search engine fundamentals like missing meta descriptions or skip heading levels. Before deploying, you can ask your agent to evaluate your page's discoverability using Lighthouse's SEO audits to ensure crawlers can effectively index your content. As for any other Lighthouse category, just prompt, "Please run a Lighthouse SEO audit on this URL and fix all findings." With the same demo site as in the previous example, Anti-Gravity is our agent found missing meta description tags across various different layouts. Easy to forget and for an LLM, just as easy to fix. By adding "Fix all findings" to the original prompt, rather than sending it as a follow-up prompt, we created a closed feedback loop this time. The agent identified issues and can jump straight into fixing them before running the same audits again to verify its fixes. For the last example, let's take a look at the new agentic browsing category in Lighthouse. Agents accessing websites rely heavily on a well-formed accessibility tree, descriptive tools, and stable layouts to navigate and interact with your website. A malformed tool or unexpected layout shift can completely break an agentic workflow. To ensure your site is ready for the agentic web, ask your coding agent to run the new agentic browsing audits. To do so, you can prompt, "Please run a Lighthouse agentic browsing audit on this URL." On this example site with a Web MCP enabled car configurator, the tool definitions look all right, and as it's a single-page demo application, it doesn't have a LLM's.txt file. I got some work to do on accessibility, though. This is how you can run Lighthouse audits with Chrome DevTools for agents and when it's useful to do so. To learn more about these and other use cases of DevTools for agents, visit our documentation at this link. Let us know in the comments which issues your coding agent was able to surface with Lighthouse. Or just leave your recommendation for your favorite real-world Lighthouse. Mine is Point Bonita in California. As always, thanks for watching and see you next time. >> [music] [music]