Unit testing seems a lot like tuning to me. There's usually a pretty straightforward definition of "correct" (A=440Hz...), but it's -- at best -- distantly related to the audience's experience of the performance.
For me, I think that a key difference might be one of "directness". That is, building things that directly relate to what the user experiences, rather than following processes which you have to trust will lead to the eventual creation of something enjoyable or valuable.
Paul Graham wrote something which, for me, does a pretty good job of capturing how things feel when you're working in this kind of way[1], and also explaining why there's an impedance mismatch between this and what larger organisations tend to be looking for.