I hate that this burns me out at all, but every job I've been at has insisted on adding some sort of data-test-id throughout the app, and I've always thought it was actively obfuscating seeing what's really happening for a perceived value add of test stability that in fact didn't matter but also actively his what was really changing over time.

Anyhow, it's been a huge morale win to see really good works like the Testing Library strongly emphasize using page accessibility hooks to drive testing! https://testing-library.com/docs/queries/about/#priority

Exactly as you say, it drives better accessibility, and gets everyone using the same referants (versus inventing a new third way outside of both query selectors & aria).

You would need the accuracy of these models to be way, way, better than today (so you do not lie to the user which has no idea about the fact that the output is a lie, because he can't compare the model output with the input, because he can't see that), their hardware requirements way, way, lower (so you don't need a game rig for useful computing in these groups), so, no, I can't agree, this A.I. approach will not save you from correct semantic markup.

> These problems are better solved through accessibility focused computer vision models.

This may be coming soon but I think it's a bad trend. Wiring app properly is much better. It's like native apps vs Electron apps, more effort = better result.

> ARIA is a 'boil the ocean' solution that primarily helps the accessibility industry

Disagree. If you are a dev capable of reading some references then you can implement ARIA. You don't need to outsource it to some separate "industry"