Google Chrome developer relations engineer Tara Agyemang explains why agents can struggle with web interfaces and recommends semantic HTML, accessibility and performance as foundational improvements. She presents WebMCP as a proposed way for sites to expose structured actions to browser agentsComputer use lets an AI agent interact with software through interface controls such as screens, buttons, keyboards and pointers..
Using maze and ticket-booking demonstrations, she compares declarative form attributes with custom JavaScript tools and emphasizes keeping the interface visible and users in control of consequential actionsHuman-in-the-loop AI keeps a person involved in reviewing, correcting, approving, or guiding an AI system's work.. She describes the API as experimental at the time of the talk.
Watch on YouTube
