Too much of AI lives in the computer without impacting patients’ or clinicians’ lives. We use randomized clinical trials and prospective cohort studies to embed AI into clinical workflows and test its impact on clinical and workflow outcomes. Much of this work involves merging AI with proven concepts from behavioral economics and implementation science. Additionally, we believe that end-user perspectives are critical to ensure that AI is collaborative and not antagonistic in healthcare. We conduct mixed-methods studies using qualitative and quantitative methods to seek clinician, patient, and other stakeholder perspectives. This work helps inform the deployment of AI in prospective applications.