This article argues that AI adoption in higher education is fundamentally an epistemology problem: different campus groups hold different assumptions about what counts as knowledge, what learning is, and what kind of “knower” AI really is—so governance fails when it treats AI as a single, stable tool. It proposes shifting from one-off policies to learning-oriented governance by convening a structured, cross-functional campus conversation that surfaces assumptions, aligns on evidence standards, and iteratively updates decisions as models, pilots, and institutional understanding evolve.