At age 38 I'm a little bit of an oldster for this industry, and I remember PCs in the 80s. I was a kid but that's when I learned to code and I distinctly remember what it was like.
It was a lot like today: loads of fragmentation and loads of vertical 'silos' with their own peculiar way of doing things.
Until the late 1980s you had a million little vertically integrated platforms: Apple II, TRS-80, Commodore 64, TI-99, IBM PC, etc. Even within manufacturers you had major incompatibilities, like Commodore VIC-20 vs. Commodore 64. Those were almost totally different machines.
Then the explosion of cheap IBM PC/AT clones changed all that and killed all the smaller players. There was a brief period from about 1988 until roughly 1999 when we had essentially one platform MS/Intel. That was MS-DOS and Windows, and the latter would usually run DOS apps. You also had one UI metaphor: the mouse and the keyboard and the screen. You had some hardware fragmentation but if you were targeting OS APIs and not trying to be bare metal it wasn't too hard to deal with. You mostly had one platform you could target and get 90%+ of the end-user market.
No more. Today if you want that 90th percentile of the user base you have to roll at least three completely different UIs at a minimum. Either that or you go all-web, which as this author correctly points out is a major headache if you want something that's mobile friendly. We've had good luck with bootstrap+react but it is more work than targeting a single native platform. (Not sure if I buy that many native platforms are more work... probably depends on the stack.)
On one hand all this diversity is interesting, but on the other hand if you just want to ship a damn product it's infuriating. It's also true on the backend: you have at least six Linux distributions, Docker, a million different 'stacks', clouds like Amazon building closed mainframe platforms like lambda, etc.