Today, many are picturing morning rounds that will soon look like this:
The team enters the patient’s room, computers and tablets at the ready. The medical student reads the patient note, differential diagnosis, and proposed plan that the EMR’s AI feature generated at admission. The supervising resident, when questioned by the attending, stumbles over an interpretation of the admission labs without first running his own thinking through AI. The traditionally trained attending catches herself wanting to run her own thinking through AI first. Then the system goes down. The patient looks up expectantly and asks, "So what's wrong with me, and what's the plan?" Three clinicians stand at the bedside, looking at dark screens, and for a long moment, nobody says anything.
It's a familiar trope. The newest technology arrives, and the fear is that we forget how to think without it. Calculators, spell-check, and search engines all engendered the same conversations, and now we are telling ourselves the same about AI in clinical education. There is a real truth here about what we stand to lose that is worth taking seriously. Deskilling. Mis-skilling. Never-skilling. Losing the tolerance for uncertainty. Losing the nerve to disagree. Losing the ability to notice what isn't being considered.
Strip those worries down and they're all the same worry. As Michelle Daniels and I recently discussed on my podcast, expertise is built through struggle, yet struggle isn't innately enjoyable. To see real results takes time, and a winding path. The human mind will happily take a shortcut whenever one is offered, and AI is the most capable shortcut anyone has ever built. That's the real concern under all of our worries, yet our concerns deserve more than vigilance. If that's the only stance we're going to take in this moment, we'll spend the next decade playing defense against a technology that's designed to win.
Instead, what if we asked bigger, more human questions: what does this moment also make possible, for the clinicians already in practice and the ones we're training right now? And what do those possibilities create for our patients and their care? That's the conversation I want to start.
Yes, vigilance is where we begin. It tells us where to plant our feet. What matters thereafter is whether we stop at what we’re guarding or we use the opportunity to expand our thinking. I propose we begin the habit of asking three higher order questions whenever the ground under clinical education shifts to this extent.
What do we need to retain? What can we now reclaim? And what do we stand to gain?
Retain comes first. AI didn't arrive developmentally, moving through the field one generation at a time. It arrived monumentally, hitting all of us at once, wherever we happened to be in our clinical career: as a student learning to build a differential, as a resident whose judgment is still forming, as an attending with twenty years of experience to draw from.
What each of us must retain is the drive to keep cognitively improving through effort, to put the pieces together in our own minds, because while AI may be able to think for us, it can't understand for us. That understanding is built one hard case at a time. Different clinical levels, same mechanism, no shortcuts. Strip that away and the fear we are naming comes true: clinicians become translators of someone else's (or something else’s) reasoning, skilled at delivering conclusions they didn't reach and can't defend.
Reclaim is the most hopeful of the three, because it allows us to pick back up what the pre-AI workflows crowded out. While many rightly assert that this is a new dawn of found time, most conversations I'm in stop there. They shouldn't. Under high workloads, cognitive rigor slips. This is the moment to rebuild those habits, before something else claims the time. Interacting well with AI can strengthen our core discipline of inquiry and evaluation. Science education taught us to observe, identify what we know and what we don’t, and then articulate a question sharp enough to answer well. Evidence-based medicine asks us to pose well framed and relevant questions, evaluate the output, and determine if it meets our context before applying it to patient care. These habits of mind are now more important than ever, and using AI safely and effectively naturally puts that discipline back into daily use. It’s the same muscle that makes us better at the bedside, better with the literature, better at teaching a student how to think. AI isn't taking this skill from us. It's insisting we strengthen it.
Gain is where this stops being defensive and starts getting interesting, because the ceiling is rising. What AI has now made possible exceeds our current clinical skills.
Consider what an AI-supported workup produces: six plausible diagnoses, four literature-supported management pathways, eleven flagged values, all of it correct, all of it fast, none of it prioritized for this patient in this bed with this family and this trajectory. The clinical skill that moment demands is triage of cognitive abundance, and that volume is new. We’ve trained generations of clinicians to find the missing piece. We now have to train them to choose among too many pieces, quickly, under pressure, without losing the thread.
Then there's the cognitive partnership itself, which is subtler. Working well alongside these tools means knowing when to let an LLM run ahead of you and when to hold it back, when its apparent confidence in its output should raise your suspicion rather than settle it, and when to use its breadth and clinical distance as a check against your own bias. We were both trained on data. Its training was scraped at scale; ours arrived one patient at a time, with a face attached and consequences that followed. Neither data set is complete alone. That's a relationship, and it has no precedent in medical education. We have never had to teach anyone how to think alongside a non-human partner that is sometimes brilliant, sometimes wrong, and rarely uncertain.
The largest and hardest gain, I think, is clinical integration: carrying a correct, yet complex, diagnosis and plan produced in seconds into shared decision-making with a patient who doesn't like to take pills and whose daughter, the one who fills the pillbox every Sunday, moves out in April. Medicine already moves faster than patients can absorb. A correct diagnosis and an elegant plan delivered overnight are not yet better care. Closing that gap takes cognitive and interpersonal work, and it takes real human time. That work has always been at the heart of good clinical care. It is about to become its central act. The better these tools get, the more they will require of us. Can we use this new reality to return to the bedside in deeper partnership with our patients?
Easier to type than to live.
With the pressures we feel already, there are days when none of this feels reachable. And underneath it, an existential question: what does prepared even mean now? Our curricula, our milestones, and our sense of what a competent graduate looks like are being called into question every day. Are we teaching and assessing against standards our field has yet to redefine?
That's where asking these three questions can help in real time. A student presents a case and reads an output as though it were an answer, and you have thirty seconds to decide what to do with that. Ask which of the three questions this teaching moment belongs to, and clarity and agency return, in the exact moment you're likely to feel you've lost them both.
Retain her need to think this through herself? Reclaim the time AI gave you back to show her how to evaluate the output? Or gain, together, the skill of understanding this complex, swift diagnosis and discussing it with her patient? Any of them is good. So are all three.
Asking these questions is a habit worth building now, because something else will come along after this. Someone will stand where we're standing and argue for protecting what AI made possible, which is exactly why the habit matters more than any single action it produces. The patient in that room is still going to look up and ask what's wrong and what the plan is. Our job is to make sure someone can still answer, and still answer well. That is the work of reimagining clinical learning.
In the next post, I want to take up what everything rests on: seeing the whole clinician take shape, rather than a stack of disconnected scores from one rotation to the next, and how AI can help us get there.
Aquifer is built by a national Consortium of clinical educators, because the educators closest to the work should be the ones creating, evaluating, and pushing it forward. The same three questions shape what we build. Retain means we protect the cognitive work that develops our students' reasoning rather than creating tools that shortcut it. Reclaim means using AI to deliver core curricular learning and individualized feedback at scale, so educators get the hours back to help students apply their learning in real clinical care. Gain means building toward what wasn't possible before, including competency-based assessment that follows a learner's whole trajectory.
To learn more and access our playbook, see Clinical Learning, Reimagined.