I could see this opening up a new way to generate ideas from which you can start building, or what to add/remove when you're stumped. Maybe a way to create coherent arrangements from a curated list of one shot samples and short loops. And I wouldn't be surprised at all if the automated mixing/mastering services (of which there has been a bunch for years) take a giant leap forward because of better technology.

But in general I'd imagine written language to be a pretty infuriating tool for describing what you want musically, when the most interesting parts of music are just about always the ones that you can't really capture with language. You can kind of outline things with written language and traditional music theory, but it's usually just a blurry version of why a specific piece of music resonates.

I think that AI tools for music will most likely just stay as plugins within the more traditional DAW structure. There's only so many ways to represent an audio file, and a fader that controls the volume of a track or some other parameter.

Like mentioned in the article, most of these additions take quite a bit away from the amount of control the artist has over the music, and lowering the amount of 'input resolution' in this sense is a block that's almost impossible to overcome.