A compiler translates _what you have already implemented_ into another computer runnable language. There is an actual grammar that defines the rules. It does not generate new business logic or assumptions. You have already done the work and taken all the decisions that needed critical thought, it's just being translated _instruction by instruction_. (btw you should check how compilers work, it's fun)

Using an LLM is more akin to copying from Stackoverflow than using a compiler/transpiler.

In the same way, I see org charts that put developers above AI managers, which are above AI developers. This is just smoke. You can't have LLMs generating thousands of lines of code independently. Unless you want a dumpster fire very quickly...

Sure, there's a huge jump from a line editor like `ed` to a screen editor like `vi` or `emacs`, but from there on, it was diminishing returns really (a good debugger was usually the biggest benefit next) — I've also had the "pleasure" of having to use `echo`, `cat` and `sed` to edit complex code in a restricted, embedded environment, and while it made iterations slower, not that much more slower than if I had a full IDE at my disposal.

In general, if I am in a good mood (and thus not annoyed at having to do so many things "manually"), I am probably only 20% slower than with my fully configured IDE at coding things up, which translates to less than 5% of slow down on actually delivering the thing I am working on.