@SteveFaulkner This is an excellent starting point for my investigation - thanks for linking this!

We’ll see what the AI wizards bring in 2026 😅

@yatil @SteveFaulkner Trusting the output is the main problem, I agree. I think if you use the AI in controlled environments for simple tests you can trust the output more, but that already reduces a bit of the value it brings.

I think deterministic testing + AI testing and then manual testing is still going to be the way to go for the near future.

@yatil @SteveFaulkner I feel that’s black and white thinking on the matter.
There is significant ability to - at least - discern things like link descriptions, image alt texts, etc.

That still won’t get us nowhere near close to 100% coverage of course, but it’s still useful to know.

@SteveFaulkner I will start 2026 by compiling a list of success criteria with a rough guesstimate of how “simple” they would be to automatically test using AI - or how “accurate” an AI result could be.

Surprisingly I haven’t found that much existing research on that, have you?