Google and Microsoft have released WebMCP - a W3C standard that allows websites to explicitly describe their actions for AI agents. Instead of the agent taking a screenshot of the page and trying to figure out where to click (slow, expensive, unreliable), the site simply publishes a "contract": here’s the function buyTicket(destination, date) - call it.
Why this matters. Currently, AI agents interact with the web like a blind person with a stick - they poke, guess, and make mistakes. WebMCP gives them a direct interface. It’s like the difference between a screen reader and an API.
But the most interesting part is the economic side. The site sees that an agent has come to it, not a human. This means it can show agents different prices, different conditions, different content. Essentially, a new class of "clients" emerges - and a separate funnel can be built for them.
Available in Chrome 145+ in early preview. Currently behind a flag, but the direction is clear - the web is being restructured for a new user who doesn’t click with a mouse.