Agentic AI and related systems in nursing and nursing-adjacent work: a scoping review of task allocation, decision authority, and accountability

Introduction Agentic artificial intelligence (AI) and related systems are being investigated for nursing and nursing-adjacent work, but inconsistent terminology and heterogeneous evaluations obscure their operational capabilities and clinical relevance. This review examined the relationship between reported agentic behavior, evaluation in nursing practice, retained decision authority, and evidence concerning patient benefit, workload, and accountability. Methods We conducted a scoping review informed by JBI guidance and reported in accordance with PRISMA-ScR. We searched PubMed, Embase, CINAHL, Scopus, and Web of Science, supplemented by citation and linked-report searching. Results Among 28 reports, 13 satisfied all criteria, four were indeterminate, and 11 lacked at least one criterion. Only two qualifying reports evaluated live nurse–AI workflows. Nursing participation ranged from clinical work to education, co-design, output rating, and outcome assessment, and did not uniformly represent runtime control. Patient-facing clinical workflows retained professional decision authority; robotics evaluations involved neither nurses nor patients. Outcomes principally concerned technical performance, communication, usability, or component-specific efficiency. The evidence did not establish autonomous nursing practice, sustained patient benefit, or reduced complete-episode nursing workload. No empirical evaluation of accountability or redress following AI-related error was identified. Discussion Operational qualification, clinical benefit, and accountable use require separate evaluation. Future studies should assess the complete human–AI pathway, including verification, correction, escalation, patient outcomes, and responses to failure. Comparative designs should isolate agentic mechanisms from additional computation, information access, and workflow redesign. A proposed post hoc , unvalidated heuristic links allocated function, decision authority, assurance, and accountability/redress. Interpretation is limited by absent prospective registration, unquantified agreement before consensus, related reports, and the absence of formal risk-of-bias assessment. Conclusion Agentic capability alone does not establish clinical readiness. Expanding AI-enabled functions requires evidence of meaningful benefit, effective human control, and accountability arrangements proportionate to the consequences of system actions.