Live coding systems are only really good for building teeny tiny little toy automata, which usually do simple stuff with note lists, and maybe apply a bit of randomness or some simple repeating functions to a parameter or five.
Weapons-grade commercial music is much more complicated. The sounds are richer, the arrangements and mixes are immensely complex (even for simple songs), and generally there's a ridiculous amount of care and detail.
But... if you start building in machine learning, database searches, and super-complex DSP patches, and the stuff you really need to make non-trivial music, you're not really live coding any more - you're plugging pre-existing modules and data together by typing.
Also, watching people typing is kind of dull.
I think there are solutions to both problems, but REPL-music systems are only ever going to be a step on the way to them.
I'd be surprised if live coding ever makes it into the mainstream in the way that (say) eSports almost have.