For example, I have friends that released a rather ho-hum mobile app. They quickly garnered something like a 1.5 star rating, scathing reviews, and almost no conversions. The business cycle on this was a year, and they are still trying to claw back reputation and win users. It's a debacle. (the problems weren't their fault, but that is irrelevant to this point).
Then you have companies with secrecy, like Apple. I think this advice would be terrible for them (I have never worked there, and am open to correction). They can't dog food it widely due to the internal silos, and they certainly cannot test it with the public.
Then there are electronic systems - iterations on SW is easy, iterations on HW expensive and hard, even with simulations, mock ups, and what have you. I worked on an augmented reality hardware thingy several years ago; we went from foam cutouts to a couple of very expensive prototypes, and that was it.
It is awesome when we can completely sidestep a problem, and this process lets you sometimes sidestep the serious difficulty of UI design. I worry when it gets bandied about as a truism, or The One True Way (not saying Jeff is doing that, I'm remarking on the wider industry). Yes, Agile lets you sidestep the problem of scheduling and estimation - sometimes. Try that when you are making a new airliner, building a cloverleaf interchange, making a car entertainment system, and so on.
edit: the converse problem is equally as large. Someone below mentioned the 'planning mania' of companies. I don't mean to downplay that problem, just to point out the need to evaluate each situation on it's particular needs as opposed to a 'best practices' (oh, how I hate that term) unthinking approach.