Artificial intelligence may help reshape patient care by handling routine clinical and administrative tasks, helping patients navigate the health system, and giving physicians more time for complex decisions and patient relationships, experts said during a panel discussion at Cleveland Clinic’s AI Summit for Healthcare Professionals.
AI could support patients before, during, and after appointments, but its expanding role raises questions about regulation, payment, and how much responsibility to give autonomous systems, panelists said.
James Gutierrez, MD, chief of Cleveland Clinic’s Primary Care Institute, moderated the discussion, asking how AI could improve care, strengthen patient relationships, and change the physician’s role. Panelists included Eric Boose, MD, a family physician and associate chief health information officer at Cleveland Clinic; Adam Oskowitz, MD, PhD, a vascular surgeon who co-founded Doctronic and serves as co-CEO; and Will Kimbrough, MD, an internist and chief medical officer of AI at Amazon One Medical.
Dr. Kimbrough said AI’s role depends partly on how primary care is defined. If primary care is the first place patients turn for health information or treatment, “AI is a natural fit,” he said, because many patients already use it to ask about symptoms and decide what to do next. In ongoing primary care, however, AI may be most useful for strengthening, rather than replacing, the relationship between physicians and patients.
AI could make visits “more valuable” by “taking some of the clutter and the distractions off,” he said.
Dr. Oskowitz said AI is particularly well suited to routine, low-risk tasks that follow clear clinical guidelines. This could give physicians more time for complex cases and conversations requiring judgment and compassion. “The tasks physicians do will change,” he said. “The more complex problems, the things where there may not be a right or wrong answer, those are the things that humans should really be doing.”
Dr. Boose pointed to prescription refills, prior authorizations, forms, and patient portal messages as examples of work that can consume physicians’ time outside visits. “This is so ready for AI to take care of these tedious tasks for us,” he said. “This is really a unique time in our careers to see such a large change in technology that’s, I think, going to make a big impact.”
Cleveland Clinic physicians are already using ambient AI software that listens during visits and drafts clinical notes. Dr. Boose said physicians initially questioned whether it could document visits accurately but were generally impressed. “Now it’s so ubiquitous, it’s kind of become almost a commodity,” he said. “Everybody’s like, ‘Well, what’s next? Because now I’ve got my tedious documentation work done. What can you do for me?’ "
Potential next steps include preparing orders, answering questions about patients’ medical records, and providing education before visits. Dr. Boose described using an AI chart tool to treat a patient who had a cat bite and roughly a dozen antibiotic allergies. Within a minute, the tool suggested two antibiotics that could be used safely.
AI could also make appointment scheduling more flexible. Dr. Boose envisioned systems that estimate how much time each patient will need, adjust schedules throughout the day, and allow physicians to contact patients by telehealth when time becomes available.
AI agents may also expand access to care outside regular office hours. Dr. Kimbrough said 68% of conversations with Amazon’s Health AI service occur after hours. The patient-facing system helps patients assess their symptoms and decide what type of care they may need. A recommendation to seek emergency care for chest pain should not be considered a failure, he said. It can reassure patients that emergency care is worth the time and cost. For less urgent symptoms, the system may advise patients to wait and contact their primary care physician.
“There’s an empowerment to the patients that we can really achieve through AI,” Dr. Kimbrough said.
Dr. Oskowitz said reducing the cost of routine care could particularly help patients without health insurance or regular access to primary care. AI could help physicians work more efficiently, see more patients, and provide care at a lower cost.
“If you leverage AI appropriately with the right safeguards, it could be almost free,” he said of some low-complexity care. “That sounds fanciful, but in reality, if you are able to deliver on some of the low-complexity care and make it accessible through AI, you can actually drive the cost down that low.”
Dr. Oskowitz said fewer physician visits could also challenge payment models that compensate practices primarily for encounters.
“As our systems have gotten better, we actually see less utilization of physicians,” he said. “A patient gets to choose when they see a physician, and sometimes they choose not to do that. That’s actually going to be something we’re going to have to think through as we go through this process of integrating AI into our system.”
Regulation must also keep pace with AI, panelists said. Doctronic is working through a Utah regulatory “sandbox” to evaluate AI-supported prescription renewals. The approach allows the company and state to test the system’s safety and effectiveness.
“The regulations we have in place today were written 25 or 30 years ago, when the concept of AI didn’t really exist,” Dr. Oskowitz said.
Dr. Kimbrough said more complex tasks require stronger safety measures. Renewing a prescription under specific conditions is less complicated than making a new diagnosis. As a result, AI will likely be used to diagnose specific conditions before it is trusted to assess a broad range of clinical problems.
At Amazon One Medical, AI is being used for clinical decision support, documentation, billing, and prior authorizations, he said, and about 80% of prior authorizations are handled automatically.
Despite those advances, panelists saw an enduring role for physicians. Dr. Boose said AI may struggle to recognize subtle changes in a patient known over many years or reproduce the empathy involved in accompanying patients through serious illness.
“The intuition and the empathy would be hard to replicate,” he said.
Dr. Kimbrough said uncertainty is another challenge for AI. Physicians sometimes do not know exactly what is causing a symptom but can still reassure patients that nothing appears seriously wrong. Large language models are “fundamentally incapable of that” at present, he said. “They’re incapable of living with the uncertainty, of taking that responsibility themselves and saying, ‘I don’t know.’”
Dr. Oskowitz predicted that AI could handle patients’ routine needs, allowing primary care physicians to care for many more patients. Physicians would continue to take the lead when cases become more complex.
“Practicing the art of healing, that’s why a lot of us got into this,” he said. “That’s something I don’t think AI will ever really replicate.”
