DRL is fun, that's what matters! :) Though classical non-biological RL has some strong assumptions (Markovian) that may not work in real world (it's nice to see fixed points with Bellman operator in theory), but for some reason with the "Deep" part, magic happens.
I disagree with the conclusion. Article has some critiques on DRL, but I don't think those invalidate the field as a whole. Even the article has a section for "When could Deep RL work for me?"