What is happening?
BrowserKitten represents a selection-first interaction: the user points at the live page and describes the desired outcome, instead of asking an agent to infer every relevant element from a text prompt.
Why people care
The interaction model narrows the problem before the agent acts. That can improve clarity, but it also makes the browser environment and the user’s existing session part of the system’s security boundary.
What to watch
Compare the reliability of selection-first and prompt-first flows, then inspect how each handles credentials, extensions, downloads and irreversible actions. A clearer interaction does not remove the need for isolation and confirmation.