A comprehensive review of generative AI in epilepsy care found strong progress in text-focused clinical extraction tasks, but concluded the technology is not ready to be trusted on its own across epilepsy decision points. The review mapped 24 studies using PRISMA-ScR methods and highlighted uneven capability—where structured information extraction can work well, specialized clinical judgments remain weak. Separately, an FDA-cleared AI decision support tool for breast ultrasound showed a path toward reducing benign biopsies by classifying many BI-RADS 4 lesions as probably benign, though the evidence is retrospective. The two stories reflect a divide: AI can improve workflow performance, but external validity and deployment-level evaluation remain the gating items for clinical adoption.