Comment by densekernel

I agree, human-steered, AI-implemented test cases can at least capture the acceptance criteria.

It's then more efficient to inspect if existing test cases are being modified as part of the delivery of something new and inspect why.