Your AI agent writes three paragraphs when one would do. It restates the question before answering it. It hedges when it should commit, apologizes when it should proceed, and thanks the caller for their input before getting to the point. None of this is a bug. It’s the intended behavior of a model optimized to make humans feel good about the interaction.
