GPT-4 can generate and iterate on simple GUI applications on the fly from text specifications. It’s primitive, but it’s the beginning of something. In the future, will it become as common to create and use ad-hoc interfaces which perfectly suit one’s immediate needs as to use expert-designed “apps”?
This is one sort-of route to Malleable software: maybe we get malleability by generating the whole app in the first place. Or maybe exposing “plugin” and “extension” layers to users becomes much more compelling when non-professional end-users can easily create those things. Or maybe open-source GUI applications will become more central: you’ll ask the LLM to modify Audacity to add X and change Y.