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.