Ah, but if you are doing groundbreaking AI research, you need AI experts. You will also need developers, but those developers may not actually need to be superstars.
In Montreal, we have tons of AI people here - their code is sometimes a nightmare. They are AI guys, 'code' to them is like 'mathlab'. Fortunately, it's surprisingly not a lot of code.
But the general point stands: it depends what you're doing.
Developing frameworks is a good example: you generally want experienced people who have 'seen it, done it, been there' and have the 'wisdom' of that - plus - hopefully, an ability to actually 'get stuff done'. Doing 'new' is hard.
But 'in general' I think the post makes sense: 95% of companies need 'professional, thoughtful, collaborators' who can take mildly complex problems, write clean, safe, simple code, document, test, release and move on ...