I think LLMs also kill off most programming languages. I think we’ll end up with a handful of languages that LLMs most proficient at writing for and the languages required for device or processor compatibility.

The cost improvement for an LLM to emit a feature (with an engineer in the loop) is too much of an improvement. We’ll look at engineers coding in C the same way we look at engineers today who code in assembly. LLM enabled development becomes the new abstraction; probably with a grammar and system for stronger specification formalization.

I've done a bit of experimentation with that. And the irony is Rust seems to fare best because of the error messages of the compiler.

I have had a lot of conversations with Claude about it and it supports that theory.