I agree with the on-demand activation mode, where the AI opens a GUI for users to interact with when necessary. However, I disagree with the idea of using the AI and then discarding it. There are situations where users and AI need to continue interacting on the GUI, especially when language is difficult to describe. For example, modifying the font in a specific location on an image, or a paragraph in an article. It's just that there's a lack of such application cases at present.