My wedding budget was tight, but my wife and I really wanted a live string quartet during the ceremony. I floated the idea of playing a recording through a speaker. That idea was shot down. Listening to an ensemble live just hits differently than hearing it through a speaker. It feels more human.
We went with the live string quartet, and I have no regrets. It was beautiful and worth the extra money.
I see a parallel with artificial intelligence (AI) and our profession. At some point, AI will become so capable of evaluating and managing patients that people may consider choosing the “digital version” of medicine over the “human version.” I see their point. But that human connection—the live string quartet—makes me think AI is unlikely to replace us and will instead make us better.
Current Pulse Check on Artificial Intelligence
AI has infiltrated all corners of patient care, helping automate routine tasks, assist with medical decision making, fight insurers, and engage patients. It’s not going away anytime soon and will only become more in-our-face. These tools, if you used them, really do speed things up and better patient care. As a trainee, I’ve used them to help find gaps in my clinical thinking and for teaching. I’ve also used it to help streamline patient care, but having it create discharge summaries, explain hospital courses to patients, write comprehensive advance-care-planning notes.
The AI tools are becoming so robust, that you can easily see how it could manage a patient by itself:
A patient inputs their symptoms, while the AI has access to their medical record data
The AI forms a differential and a management plan
The AI pends orders and next steps for the physician to sign
We’re probably already at this point. Counsel Health is doing it.
While many fear this is how AI will take our clinical jobs, I fear (in a good way) that this is how AI will make us busier and more in-demand!
We’re Not Going Anywhere
Dr. Dhruv Khullar, one of my favorite writers and a frequent contributor to The New Yorker, explored this in his NEJM piece, Artificial Intelligence and the Future of the Clinical Workforce. Drawing on economic theory and history, he argues that AI agents could expand the clinical workforce over time. He offers three main reasons:
Jevons Paradox: making a resource more efficient to use can increase overall demand for it. More efficient steam engines burned less coal individually, but their adoption drove demand for more engines and, ultimately, more coal. Something similar could happen in medicine. If AI helps triage patients with asthma, cough, or COPD and prepares notes and orders for our review, we may be able to see more patients in the same amount of time. In the ICU, drafting notes, preparing orders, and flagging safety concerns—such as prolonged intubation or an indwelling Foley catheter that needs reassessment—could free up time for patient care. If those efficiencies make care more accessible, demand for our services could grow.
Lump of Labor Fallacy: the amount of work in medicine is not fixed. Technology can create new work by making treatments and services possible that didn’t exist before. Physicians in the early 1900s would have struggled to imagine the radiology capabilities we rely on today. I could imagine AI creating a new niche in physiology, whether in the ICU or outpatient practice. We may find ourselves doing work we haven’t yet thought to train for.
Tasks vs Skills: automating individual tasks doesn’t necessarily automate the clinician’s job. Being skillful means bringing together information from different sources, interpreting labs, weighing risks, managing uncertainty, adjusting plans, and working with colleagues to make decisions. These activities are interconnected, and automating some could make the remaining human contributions more valuable. AI’s ability to draft a note or analyze a test doesn’t, by itself, establish that it can take over the full job.
Dashevsky’s Dissection
AI will make many parts of our jobs easier while creating more demand for our time. I’m optimistic about that, especially if it means we can spend more of the day caring for patients and less of it working through everything required to deliver that care.
Think about the workflow from a new patient appointment through testing, treatment, and follow-up. AI could help prepare the chart, draft the note, organize next steps, and explain the plan to the patient. Each task takes time. If we can complete them more efficiently, we may be able to see patients sooner and follow up more consistently. Patients who previously struggled to get care may finally have a way in.
That’s where Khullar’s argument makes sense to me. We already have patients waiting for our services. Making those services easier to deliver could help us meet that demand, while new capabilities could create additional (billable) work. I can imagine our schedules staying full even as the work within each visit changes.
In the ICU, I’d be careful about assuming that faster workflows automatically mean faster patient turnover. Patients still need time to recover, and discharge depends on much more than how quickly we finish a note or place an order. But reducing those delays could give us more time to reassess a patient, talk with a family, or work through a difficult decision with the team.
I picture it as skating on fresh ice after the Zamboni has passed. We still have to choose the direction and make adjustments, but moving forward takes less effort. Our current workflows can feel closer to mountain biking: we’re trying to reach a destination while navigating obstacles and repeatedly getting back on track. AI could smooth out some of that terrain, even though the clinical uncertainty remains.
And I keep coming back to the string quartet at my wedding. We valued having people there, sharing that moment with us, enough to pay more for it. I think many patients will continue to value a clinician who can sit with them, explain what we know, acknowledge what we don’t, and help them decide what to do next. My hope is that AI gives us more room for that part of medicine as demand for our care grows.
