My question is why use the web at all? Surely there's an easier way for a blind person to access an information or send data to an API than through an HTML web page.
Oh wait, except most of the audience isn't blind, and starts using clients that can do ever-zanier things with the UI. Designers want to appeal to this market. The beautiful, parseable structure of HTML gets forgotten about, and it's no longer usable by the blind.
So you want content providers to support an api endpoint that works like HTML used to? Okay, but what stops that exact process from happening all over again?
Or, maybe you mean that every provider in existence will support two versions of every endpoint: the meth-addled zany design, and the standards-compliant, usable one. Okay, that could work, but it seems a lot harder than just having one version that makes a small effort to make itself more machine parseable.
But if you're doing anything that spits out HTML (flask, django, react, vue, whatever), you should be able to construct that HTML in an accessible way.
Too many web developers go for the full JS framework-heavy application architecture without stopping to consider if the project in question is trying to replace a native desktop app or if it's trying to replace a PDF document that could be printed on paper. The latter projects need to be pushed back toward simpler HTML.
From where I was sitting it seemed more like "brands demand 'consistent brand image', which they confuse with pixel perfect design because they don't really grok the new paradigm, and so any semblance of semantic web and accessibility dies".
And we've just now 15 or more years on clawed our way back to some semantics, and responsive design, and a legal framework that forces brands to accommodate those using different UA for accessibility reasons.
I would also reconsider why that “API” would be better than the existing web APIs which the community has been working on for the last few decades. HTML is great for this purpose: you have a rich document markup language with a decent semantic model, separation of display styling from content, and the ability to add metadata to support different accessibility modes. Using it to build applications is problematic but still possible: the main problem being that many JavaScript tools are developed by teams which don't prioritize accessibility, which is really the same problem as the previous paragraph where things which the people building the site don't check tend not to work. That's getting better but it really needs a combination of awareness of the need and legal ramifications to get people to build it in from the beginning.