Syntax: give browser agents useful tasks, not a button for every click
On Syntax, Chrome’s Sarah Drasner and Dominic Farolino explain how WebMCP lets a website expose application actions directly to agents. Instead of repeatedly interpreting screenshots and clicking through menus, an agent can invoke a meaningful task. The discussion emphasizes keeping the existing interface available so people can collaborate with the agent and see the results.
Syntax · Sarah Drasner and Dominic Farolino · Sep 9, 2026
Why it matters
Useful for web developers deciding what capabilities an agent should access in an existing application.
More context
This card covers the transcript from roughly 01:23 to 11:14: the browser-side model, joint human-agent use, and task-level tool design. Start there for the concept rather than treating the whole interview as reviewed. The guests describe an evolving API and origin-trial feedback.
Keep in mind
This is an explanatory interview, not a measured performance comparison. Do not infer universal browser support or a finalized standard from the episode title.
Transcript read