If you continue the live coding (thought) experiment, you would naturally code a better interface. At that point you would be reinventing something like Ableton Live, but completely different. Similar to how using Lisp for a project develops its own DSL for the application different from other DSLs for other applications.
So we'd be watching someone simultaneously live code/play music and live code their interface.
Personally, I'd rather let them code the interface off stage and watch the live performance. After all, using the interface is live coding, but using the pre-built DSL.
In some ways, sonic Pi is already a higher order interface than whatever language it's written in.
I am mostly talking about the way of inputting the code though. Having someone bent over a laptop is fairly disgraceful.. Maybe people said the same when knob-turning DJs appeared.
This is a good 1-'man'-show[0] TEDx, where watching someone program/fiddle with (software and) knobs is great to watch. Talk's good too--build-up to performance/demo is about halfway through.
So we'd be watching someone simultaneously live code/play music and live code their interface.
Personally, I'd rather let them code the interface off stage and watch the live performance. After all, using the interface is live coding, but using the pre-built DSL.