AI promises to enhance clinical decision-making, improve efficiency, and support better outcomes, but realizing these benefits requires more than optimizing algorithmic performance. The impact of AI hinges on its alignment with the complexity of the clinical work it aims to support. This lecture synthesizes lessons learned from a series of applied AI projects spanning emergency care clinical decision support (CDS), patient placement, and patient-facing communication and introduces an AI-CDS toolkit for designing and deploying human-centered AI. Together, these examples underscore the need for thoughtful sociotechnical system design—and for building health systems’ capacities to evaluate, adapt, and implement AI in ways that remain accountable to both clinical realities and patient values.