Indeed, I think literate programming might see a resurgence with the help of LLM tooling. Both Copilot and ChatGPT already have some ability with it and in the former case it helps with the loss of autocomplete you typically get when doing literate programming.

The autocomplete problem is pretty easily solved by implementing custom language servers that just dispatch to two existing language servers depending on context. I was half way through doing that before realizing that the entire project was a massive yak-shave off of the thesis I needed to finish.

I do think that it should be considered best practice to leave the prompt that generated some functional code inline in a comment or so forth. It acts as a marker letting the reader know "this is what the code was intended to do". If I find some weird code on a bug hunt, I need to spend time triaging it between "weird code required because of weird intent", "weird code that works, but a simpler solution exists", and "weird code that contains the bug I'm looking for". Context helps, and if you're writing the context anyway, there's no reason to exclude it from the source.

nvim allows more than one language server to be attached to buffer...