Abstract
This paper examines the gap between what nurses need and what actually gets built, through a three-month ethnographic study at Mirae Hospital (MH) in Seoul, South Korea, combining interviews, ward observations, and participatory design workshops with hospital staff and union nurses. Our findings reveal three persistent patterns: technologies that succeeded eliminated acknowledged problems through collaborative design; technologies that failed attempted to model volitional human behavior or assumed laboratory conditions that clinical environments cannot provide; and high-impact automations that nurses explicitly requested were never developed, displaced by technically sophisticated investments aligned with institutional prestige rather than frontline need. We further show that AI adoption differs systematically between unionized and non-unionized hospitals, with union representation playing a meaningful role in ensuring AI serves workers rather than institutions. Together, these findings point to a structural problem: bedside nurses – those most directly affected by clinical AI – remain least likely to shape what gets built. Addressing this requires not only better design methods, but institutional reform in how technology priorities are set and whose needs are treated as authoritative.