I have to admit LLMs are actually quite useful at generating code for me, but I am experienced enough to know what I want. I use it as a next-generation autocomplete.

This is the "whole thing" in a nutshell for me too. It's useful, speeds some things up, but I've been doing this a long time.

The state-of-the-art or medium-term future of the tooling doesn't feel apocalyptic in itself, but the macro forces, implications of scaling, and general reactions to it on all sides are a different story.