The author probably underestimates the limitations of Unity, especially earlier versions from back when The Witness was made. It would kind of be surprising if all the things in that game were possible to do with reasonable performance in Unity back then, especially without paying ~50k+ for source access.

It feels like the article almost but doesn't quite touch on it: there's people firmly in multiple camps, and gamedev tends to be camp 3, but also very often camp 2 and/or camp 1.

To get AAA or even AA games running requires a lot of dogma from camp 2. Graphics programming, procedural generation, and gameplay systems complex enough to produce emergent gameplay would be camp 1 from the sounds of it.

I feel my feet firmly stuck in all 3, at any rate.