One trouble I have with LLM generated code is that it can be a big hassle to cut and paste multiple revisions of it into a browser window and often it is better to ask it for specific advice about a line of code here or there manually, particularly when generations are slow.

It particularly bugs me with the Jetbrains assistant which can see your source code; somehow I find it much more exhausting to cut-and-paste a few revisions of A.I. generated code which is often slightly off as opposed to cutting and pasting from stack overflow.