If there is one thing I believe that will come out of the whole AI bubble, is this: we will interact with applications though natural language and the "graphical UI" will be mostly about visual feedback of what operation is happening. The challenge for open source projects will be to define whether they will assume that their application will rely on some external LLM that will (hopefully) be good enough or if they will have to distill/train a specific model that can be smart enough to interact with the application tools.