Lower bar to start kernel development?

One of the most promising developments I see for code LLMs is the switch to a more diffusion style. So rather than generating a sequence of words, it's working on a grid of characters like your editor and iterating over that in the same way that stable-diffusion produces images. This means it has a chance of "seeing" the code as a whole and making changes globally. And so noticing mismatches between different parts and aligning them in the next iteration. So more like how people code.

The step after that would be an LLM iterating over an AST so it doesn't have to worry about getting the syntax right, but I haven't read about that yet. It's not clear to me if that technology even exists yet.