Agentic AI in Healthcare: When Autonomy Needs Guardrails
Scenario: Eating Disorder Screening in Primary Care
A primary care clinic implements an AI agent that reviews patient questionnaires before appointments. When a patient screens positive on the SCOFF questionnaire, the AI can identify potential concerns and support clinical workflows.
The question is not whether AI can detect risk—it is how much autonomy it should have once risk is identified.
Appropriate AI Actions
- Flag charts for clinician review
- Recommend follow-up assessment questions
- Provide evidence-based educational resources
- Suggest referral options and care pathways
Where Human Judgment Matters
Eating disorder screening involves complex clinical, psychological, and social factors that extend beyond a questionnaire score. A positive screen may reflect body image concerns, trauma, anxiety, depression, or varying levels of readiness for treatment.
While AI can identify potential risk and recommend next steps, decisions involving referrals, diagnoses, treatment planning, or direct patient communication should remain under clinician oversight. Human judgment provides the context needed to interpret results, build trust, and ensure patient-centered care.
Principles for Responsible Agentic AI
Human-in-the-Loop
AI supports clinical workflows, while clinicians retain responsibility for diagnosis, treatment decisions, and sensitive patient conversations.
Risk-Based Autonomy
Low-risk tasks such as chart flagging and information gathering may be automated. High-risk clinical decisions should require clinician review and approval.
Context Over Confidence
Even highly accurate AI recommendations must be evaluated within the patient’s medical history, psychosocial circumstances, and individual preferences.
Key Takeaway
The goal of agentic AI is not to replace clinical judgment but to extend clinician capacity. Effective AI governance balances efficiency with accountability, patient safety, and trust.
Faith in AI Health 