Esteban Zavaleta-Monestel, José A Castro-Gamboa, Jeaustin Mora-Jiménez
Agentic clinical artificial intelligence (AI) is moving beyond conversational assistance towards systems capable of generating medication-related recommendations and structured clinical actions. Although recent models have shown strong performance in simulated disease-management and medication-reasoning tasks, benchmark performance does not establish the safety of autonomous prescribing. Medication use depends on patient-specific information, local formularies, medicine availability, monitoring capacity, clinical workflows, and professional accountability. This editorial argues that medication-related AI should be governed according to clinical risk. Lower-risk functions may support medication reconciliation, monitoring and patient communication, whereas treatment initiation, dose adjustment, antimicrobial selection, therapeutic substitution and deprescribing should require explicit pharmacist and prescriber validation. Local prospective evaluation, transparent documentation, version control, incident monitoring, and sufficient professional authority are essential. Agentic AI may offer meaningful clinical support in medication-related workflows, but its role in prescribing should remain pharmacist-governed, locally validated, and auditable rather than autonomous.