I find that it sloppily goes back and forth between old and new methods, and as your LLM spaghetti code grows it becomes incapable of precision adding functions without breaking existing logic. All those tech demos of it instantly creating a whole app with one or a few prompts are junk. If you don't know what you're doing then as you keep adding features it WILL constantly switch up the way you make api calls(here's a file with 3 native fetch functions, let's install and use axios for no reason), the way you handle state, change your css library, etc.

{/* rest of your functions here*} - DELETED

After a while it's only safe for doing tedious things like loops and switches.

So I guess our jobs are safe for a little while longer