1) Hack together assets as close to AAA as they can get, on top of a POS tech demo. Enough to show to people and get sales (this is a very Hacker News startup-y technique)
2) Pick one part of the system that "needs to work" by release, hire a few good coders and have them actually write a proper software subsystem
3) Hack together another POS tech demo, with the proper subsystem in place
4) Redo all of the assets that have already been done to use that new subsystem
They've gone through this several times with many many subsystems. It's not clear to me this will ever crystalize into a system that has a certain quality bar it hits across the board. So in that sense you might be right that they should've taken a more top-down high falootin' architecture approach from the beginning.
But that's speculation too. Design is hard, and you generally have to try stuff in order to figure out a large system. Some degree of iteration is necessary. At a certain size waterfall just won't work.
But for sure your characterization of "he's just throwing artists at cryengine" is not correct.
I can sum up my point very quickly.. They used CryEngine.. Maybe the worst possible choice imaginable for this kind of game. They needed someone early on to choose not Cryengine, to choose some tech-stack that they wouldnt be fighting a constant losing battle against.
Instead he just seems to go on a mad hiring spree of artists, throwing more and more complete AAA assets on top of the POS visual demo he had made in Cryengine.. What boggles my mind is that it seems they never veered from this strategy, 100's of millions of dollars and 8+ years later, and it appears they are still throwing stuff willy nilly onto that original dumpster fire.
Seems like one of the best examples of the sunken cost fallacy in action that Ive ever seen.