I feel like I have good LLM tooling in my workflow for brainstorming and proofs of concept but I tend to push LLM’s for complex but not crazy audio stuff (midi and audio signal routing, automation and playback) to brainstorm getting the same functionality out of hardware with fewer cables. Everything I do is absolutely available online and in pdf manuals but the models get 30% of it wrong every time. It’s to the point where (like code) I just can’t use LLMs for anything remotely _hard_.