Or should I say, I've not written a single line of Zig because I've been managing AI's coding in Zig.
Turns out Zig is a fantastic language to "centaur" on. (Reference is to "centaur chess" and which is also sort of becoming a term indicating close code design cooperation with an LLM.)
All of that C code that the LLM trained on ends up helping it work out algorithms in Zig, except that Zig has waaaay more safety guarantees and checks. And is often faster compiled code than the equivalent C, which is astonishing.
The same can be said about Zig's comptime. It's entirely unlike anything C, C++ or Rust has to offer.
> I expect LLMs to be really good at converting C to Zig.
While it's possible to translate C to Zig code - and you don't need an LLM for that, it's a Zig compiler/build-system feature - the result will be quite different from a project that's developed in Zig from the ground up since the translation output wouldn't make use of Zig's unique features (and Zig isn't really unique as 'C translation target', C can also be translated to unsafe Rust, or even to Javascript - see early Emscripten versions).
Also, the 'C compatibility' of Zig is implemented via a separate compiler frontend, Rust toolchains could do exactly the same thing by integrating the Clang frontend into the Rust compiler.
OTOH going from C++ (OO) to Rust (not OO, borrow checker) is a big leap.