We're very much at the "Ah, but what if it *didn't* have to be reliable?" stage of this AI journey.