Friday discussion question: when was the last time you adopted an AI tool because it "seemed useful" rather than because someone showed you evidence it actually worked?

I am asking because today's blog post argues that we hold AI education tools to a much lower evidence bar than we would ever accept for, say, a new clinical intervention or a new manufacturing process. The bigger the claim a tool makes, the stronger the evidence should be. Most AI ed pitches do not survive that question.

I wrote up a 4-question filter you can run on any tool, vendor deck, or "this changed everything" testimonial:
https://www.harvestkernel.com/evidence-test-ai-education-hype/

What is one AI tool you ALMOST adopted but did not, because something about the evidence did not add up? Or one you adopted and later regretted? Either direction is useful.