Some of the demand will disappear, when, instead of adding a dependency on an open source package, to get that function you want, you use your LLM "assistant".
(The assistant will effectively plagiarize open source code, and then you don't have a dependency, and you slap your own copyright notice on it.)
LLM-generated/laundered code could move code bases a little bit back in the direction of low-dependency monoliths, like a lot of software was before the wild popularity of open source code (and, especially, modern Web dev).