Agents use browser tabs and move like a hand
Agents list, open, switch and close browser tabs, see controls inside web components, click along a curved path, and drag on the Mac.
An agent browsing for you now works with tabs: it lists what is open, opens a new tab with a page, switches between them and closes one, and gets a fresh view of the page in front each time. Closing a tab still asks first.
Pages built from web components keep their buttons inside a hidden layer that agents could not see. They can now read and press those controls like any other.
Clicks travel to their target along a short curve with a natural press and release, so sites that watch the pointer treat the agent like a person. On the Mac, an agent can drag from one point to another inside the app it is using, for sliders, files and selections.