Google published new Lighthouse documentation: experimental agentic browsing scoring

• Evaluates how well your site is built for machine integration
• Because agentic web standards are still forming, it's meant to provide actionable signals more than a definitive ranking score
• Audit results can fluctuate due to timing of tool registration & agentic request response

The stuff it checks for actually makes sense, even to a non-developer like me:

• All interactive elements have programmatic names
• Roles & parent/child relationships are valid
• Content is not hidden
• Visual stability is critical so keep an eye on CLS
• Checks for llms.txt at domain root (even tho Google's latest blog says llms.txt is a myth, this is about agent-centric accessibility, not optimizing for AI search visibility)

It offers 3 pieces of advice:

• Use WebMCP API to make your site's logic & forms clear to AI agents
• Prioritize semantic HTML & proper ARIA labeling: "these are the machine-eye view of your page"
• Optimize for stability by reducing layout shifts so agents can interact with your UI