I think the issue is that you're framing alignment in the context of an individual action, "Are we aligned in this task?", which, I agree, devolves into a rules

I think the issue is that you're framing alignment in the context of an individual action, "Are we aligned in this task?", which, I agree, devolves into a rules-based approach. I think the notion of Alignment is actually the answer, but it's (to use your framing), better described as "Meta alignment" or said differently, "alignment with a user's (or systems) long-term goals, values, and beliefs. All post-AI problem spaces are too complicated for task or even epic-level framing. Everything needs to be treated as long-running, meta-alignment frameworks.