The prevailing discourse on Artificial Intelligence (AI) use cases remains anchored to a substitution paradigm: AI
systems are evaluated primarily by the degree to which they replicate or replace pre-existing human tasks. This paper argues that such a framing is structurally insufficient and proposes a foundational reorientation. Inferential Leverage (IL) is introduced as a unifying theoretical construct defined as the ratio of reliable inferences available to a human decision-maker with AI assistance to those available without it and is used to build a novel five-class taxonomy of AI use cases organized by the nature of the epistemic transformation they produce, rather than by industry sector or technical modality. Drawing on cross-domain evidence spanning clinical medicine, environmental governance, legal infrastructure, creative cognition, and adaptive education, it is demonstrated that the most transformative AI deployments are not those that act autonomously, but those that structurally reshape the epistemic conditions under which human judgment operates. A formal IL scoring model is further developed, an algorithm for IL-aware system evaluation is presented, and design and governance principles are derived from the framework. The results suggest that AI policy, procurement, and engineering practice will be significantly improved by reorienting evaluation criteria from standalone system accuracy toward measurable inferential leverage for human decision-makers.