There is more value in small, safe, well-governed AI interventions than in flashy experiments that frequently fail. This article argues for a “boring AI” approach.