The article critiques the traditional focus on achieving consensus in multi-agent systems, suggesting that disagreement can be a valuable signal of normative uncertainty. It proposes a new framework for content moderation that utilizes reasoning-trace disagreement to enhance strategic decision-making in AI collaborations.