As a part-time software developer, one of the problems I think is the following: AI/LLM coding may actually be improving some software, true. But I think it's often in domains where we've already made a lot of progress, and AI helps put the finishing organizing touches on things.... but when it comes to new ideas or new software, I think AI is making things worse.
For instance: an AI-assisted developer can probably make a pretty decent basic text editor, but let's say they need to make an entirely new type of app, then I think the new app will lack polish or finesse because no one has made that before. Or AI can help find bugs in mature, well-written code but what if someone needs to write a new kind of app? Then there's a lack of polish again...
I guess the tendency is the problems shifting to higher levels, while for lower level or common problems things are getting more robust and better.