There's always an element of "one step forward, two steps back" with major technology shifts. Every time we've had a big change in platforms, we also had to redo existing engineering work.

I don't think this is doom and gloom for accessibility, though. The future is in general-purpose assistance technologies that mediate any application. You can smell it with the new work in ML. It is not here now, but as with everything in technology, by the time it's mature and widely available, it's nearly obsolete.