I like to think of ecosystems in discussions like this. Yes, the Google/Carnegie Hall ecosystem requires the talents of rockstars. But the places I work are a totally different ecosystem and require a different type of individual.
One other factor is that most systems and programs of a certain size are actually quite hard to work with - the "average" may not be enough, driving costs and unnecessary complexity until the system spirals into it's own collapse and eventual replacement.
One problem is that while anyone can tell if someone is any good at playing the piano, it will probably take 6 month or more to discover who is 30x more productive at developing and maintaining software, or who will in fact destroy more value than they contribute in the long run.
It's safe to say that developing software requires more than the average level of "talent" that is available in the population, and possibly also among the profession.
The other point is that the "concert pianist" and "master painter" are damaging stereotypes for the development industry--almost no one is Mozart, and the persistent and toxic idea that every programmer needs to be Mozart prevents good people from feeling good about their work.
Additionally, he makes "good programming" a bit more nuanced. "Programming" isn't a monolith that you can either be good at or bad at--it's composed of many subcomponents that you can have various competencies in without suddenly passing a magical line from good to bad. I think the argument is quite a sane one--the stereotypes that people fall back on are just that: an attempt to reduce a complex set of conditions into a simple binary.