I think important paradigm shifts can often look like this - there's not necessarily a reason to expect them to be instantly optimal. Deep Learning vs 'good old-fashioned AI' is another example of this dichotomy; it took a long time for deep learning to establish itself.