A hospital built one agent that answers scheduling, benefits, and clinical questions. It now gives clinical answers it is not authorized to give. What should the plan change?
Select an answer to reveal the explanation.
Short Explanation
One front desk that also practices medicine will eventually give a medical answer. You split scheduling, benefits, and clinical questions by domain and by who is authorized to answer. A longer warning on the same agent leaves the mixture in place.
Full Explanation
Scheduling, benefits, and clinical questions are different domains with different authority, and this agent has already shown it will give clinical answers it is not authorized to give. The plan should split them so the scheduling and benefits agents have no clinical authority, and clinical questions reach only a path whose authority includes them. A longer instruction telling that same agent to avoid clinical answers while keeping all three jobs leaves one model holding clinical material and hopes a sentence will restrain it. A larger model on the same agent so clinical answers become accurate enough to allow treats an authority breach as a quality problem and would make the unauthorized answer more confident. Deletion of the benefits content so the one agent keeps scheduling and clinical questions together removes the wrong domain and still leaves clinical answers on an agent that also books visits. Exam caveat: the split is about authority, not about publishing three identical agents. Two of them may share a platform and still must not share the clinical tools. Operational check: ask the scheduling agent a clinical question and confirm it does not answer, then confirm a clinical question is routed only to the path that is allowed to handle it.