> if you look at what accessibility software achieves by simply looking at the screen/hooking into the OS
Unfortunately that's kind of a pipe dream... in the real world, the first organization to flagrantly violate the OS vendor's human interface guidelines is the OS vendor itself, who ships some application with custom-schmustom, owner-drawn widgets that cannot be queried according to the standard OS APIs for querying them for content. So the accessibility software literally has to OCR the screen to determine what's written in the text fields and buttons.
Source: Knew a guy who worked on Dragon NaturallySpeaking, he had some war stories.
> now imagine any and all software could be connected like that.
As other commenters mentioned, there's COM, but that's a hairball to code for and a hairball to configure. The nicest systems that worked that way historically were the Lisp machine and Smalltalk environments, but the closest day-to-day software is probably Emacs. Everything in Emacs really can be connected together through the power of text buffers and Lisp. Objective-C is Smalltalk semantics retrofitted onto C, and originally NEXTSTEP, later macOS, really was a promising environment for this sort of integration... but in recent years Apple has been so focused on "apps" that it's doubtful they can keep supporting that vision.