If that future came to fruition we'd stop hearing "it works on my machine", and start hearing "it worked with my LLM".

The problem with refactoring code, re-writing codebases, and such major work is not the effort of re-coding. It is that you lose the code that has been battle-tested in production over years of use, millions of man-hours of beating on the code and turning it into a product that has survived every real-world edge case, attack, and scaling concern throughout the entire history of the product.

When you re-write code for for the sake of re-writing code, you throw that all out and have a brand new codebase that needs to go through all the production pain all over again.

So no - the trend I'm hearing of people thinking code will just become an output of an LLM-driven build process sounds quite naive to me.

You should always distribute AST-level refacs, which are deterministic, instead of prompts. You can easily prompt these out, but nothing really changes here besides who writes the migration (human vs LLM).

This is an essay, not a conversation starter. You aren't asking any questions. Maybe you should try blogging.