Making an agent call a tool to manipulate a UI does feel like normal application development and an event driven interaction... I get that.
What else drives your preference?