I have had good success using AI for search applications. In my experience this is an area where AI engines completely leave search engines in the dust _if_ the question you have is poorly phrased (you don't know the right term, or the term is used in many contexts) or when the thing you're looking for is exceedingly localized (why is there a _second_ lower case "L" in the mangled name of this variable). These kind of questions where the context is an integral part of your questions seem to be a very poor match for our current search engines and are instead answered quite usefully by the AI chatbots (they'll take a couple tries, often, but if you keep them on a leash, they tend to turn a 30minute search into a 2 minutes search plus 10 minutes of verifying what they are saying, still twice as fast). I am not clear that there is enough of a corpus of material covering lilypond that this would work to good fruition, but it seems to me that even comparatively niche communities like the TeX/LaTeX/etc people are starting to get good use from AI miles.
At this point I have not used AI to type code on my behalf, except for one example. In my case I had a piece of moderately obscure linux syscall/ioctl code, and needed the windows equivalent for it. I got a perfectly reasonable 5-liner from the chatbot. Now I can show that to a Windows person for sanity checking, but I'm off the ground in _seconds_ and all I need to do is put down a comment next to it flagging it for expert review when I send it upstream. -- Luca Fascione
