Aquifer Blog

What Duck Mode Taught Me About Feedback

Written by Leslie Fall, MD | November 18, 2026

I remember the day I learned about "duck mode." I was a second-year resident in the pediatric intensive care unit, and I had just finished running my first code. I was equally terrified and exhilarated. My patient had survived, and so had I.

Afterward, the highly experienced charge nurse (you know the one) drew me into an empty patient room, looked me in the eye, and said, “I want to give you some feedback.” (Uh oh.) “You clearly know your stuff, and you ran a successful code. Congratulations and . . . (wait for it) we are all now thoroughly exhausted.” I had followed the checklist and called out the right orders. I had also flung my anxiety around the room like a toddler with a loaded paintbrush. “Remember,” she said, “your job is to stay in duck mode: calm on the surface, and paddling like hell underwater. Do that next time.” She smiled at me and walked out.

I learned only later that she did this for every resident after their first code.

In less than a minute, she did what we have spent decades of faculty development trying to teach. Her feedback was immediate, private, and specific. She named what I did well before what I needed to change, and she gave me one thing to do next time. Many of us grew up with exactly this kind of feedback in a music lesson or on a practice field, from a teacher or coach who cared whether we got better. Then we entered clinical training, and this kind of feedback became the exception. In one study, more than half of medical students reported rarely receiving feedback from clinical teachers, and fewer than 20 percent had received it during a patient encounter.1

Feedback is the most important instructional act in clinical education and the most human one, and we have given it far less than its due.

What has changed is that we can now make it routine, the way other apprenticeships always have.

 

We built a system that makes feedback feel bad

Feedback is a relationship act, and it runs on care. When I care about a learner, and about the patients in front of us and the ones he will care for long after he leaves my rotation, telling him what I saw is one of the ways I show it. When he cares about the same things, he hears it that way. Care earns gratitude, and grace on both sides: for the learner who stumbled, and for the teacher whose words came out imperfectly, but who cared enough to try. Care is what holds our golden triangle of teacher, learner, and patient together.

However, we have built a clinical education system that engenders bad feedback, and makes it feel bad. Being at a novice stage in clinical learning is undersupported and often implicitly ridiculed or punished, making the need for feedback feel like a confession of weakness. Our faculty trained in that same system. Many never experienced frequent, well-delivered feedback themselves, and they carry the discomfort of feedback poorly given into their own teaching, often with good reason.

We responded with the best intentions: models, rules, and training designed to systematize the process, develop the faculty, and protect the learner. But over-engineered systems with impossible execution still leave everyone feeling bad. Faculty, unsure they can meet the standard, put feedback off until they must deliver it, and then it comes out as a brain dump onto a learner who can only absorb so much at once. Meanwhile, silence teaches its own lessons. Some learners grow overconfident: If I were doing something wrong, they’d tell me. Others grow anxious: When will I be hit with it all?

Adequate and frequent beats polished and rare, and adequate is far better than we think.

 

The norm, not the legend

Our field has worked to improve feedback for a generation, and the experience of giving and receiving it is still inconsistent. None of what I have said so far is new, and an AI wrapper on an old conversation will not change that. The charge nurse needed no technology at all. What we have never had are the conditions that would make her the norm rather than the legend: learners who expect feedback as part of the fabric of their learning, and teachers prepared and supported to give it.

Two developments are changing that. One elevates the importance of feedback. The other makes it easier to improve.

The stakes are rising. Competency-based medical education relies on watching how a learner develops over time and making explicit what we see. That makes feedback the mechanism the whole framework depends on. AI raises the stakes further. Experienced clinicians risk becoming deskilled and learners risk never becoming skilled,2 and the feedback that passes between them is part of how both stay sharp. AI makes the appearance of competence easier to produce, but performance does not necessarily equal understanding.3 A learner whose "faking it until he makes it" goes unnoticed may never make it, and feedback from someone watching closely is how we reach him in time.

The means are arriving. Longitudinal assessment systems can hold a learner's journey and goals in view, and AI can offer well-designed feedback more often than any faculty member could. In AI-supported simulation, a learner receives structured, specific, timely feedback, privately and with no hidden agenda. His ego gets used to it. He sees, again and again, what feedback done well looks and feels like, so that when a person offers it, it is already routine.

 

What the charge nurse knew

A skeptical reader might push back: AI is beginning to review real patient care, from our clinical documentation to the decisions we make, and it may soon offer feedback from ambient listening to the encounter itself. Why not let it carry the load? That future is coming, and we should welcome it. But think again about that charge nurse. She understood how my anxiety affected the team, recognized what I needed to learn from that moment, and knew how to deliver a message I still remember. Her feedback drew on clinical judgment, professional responsibility, and a relationship grounded in our shared commitment to the patient. Routine AI feedback prepares both learner and teacher for that deeper work. It cannot do that work for them.

 
 

Building the conditions

It starts with three ground rules.

The educators closest to the work still define what good feedback looks like. System prompts are only as good as the humans who write them. Good routine feedback also stays formative. Scored and tracked for the learner alone, it can be more useful than comments by themselves. Once those scores become visible to the clerkship director or the school, the feedback stops being formative and becomes another test to beat.

Faculty development must be a priority. Learners who arrive at the bedside having already experienced plenty of routine feedback will be less afraid of it and more ready to use it. That frees our preceptors' precious donated time for the moments that matter most, but preparing learners to receive feedback is only half the work. With AI carrying that load, faculty development can focus on reading a real moment in its full human context, choosing the one thing the learner most needs to hear, and saying it in the way it will land best. Institutions must make room for this with time and training, so feedback becomes something our faculty feel prepared to do rather than one more expectation they struggle to meet. Some joy may even come back into teaching.

Keep it simple, frequent, and spot-on. Ask the learner what he wants to work on. Watch with a clinician's studied eyes and ears. Find a private corner, offer one or two clean, fair observations, frame them as “next time,” and make sure he heard what you meant. Do it often, not just at the end of the rotation.4 As longitudinal assessment systems come online, add one step after the first: look up what he needs that you can provide. Done this way, we work through our own reluctance, our learners learn these skills by experiencing them, and together we carry forward the benefits to our patients.

 

From the exception to the expectation

The charge nurse had my back that day. She saw what I needed, told me what mattered, and gave me the confidence to do better, and she did it for every resident. Every learner deserves that kind of feedback, and our clinical teachers deserve the support to do it well.

 

How Aquifer Is Approaching This 

At Aquifer, our Consortium of more than 100 clinical educators sets the standard for what good feedback looks like, and our AI-supported formative feedback tools are built to that standard. In Virtual Patient Encounters (VPEs), learners interview an AI-powered patient and get immediate, rubric-based feedback on how they took the history: breadth, depth, clinical relevance, questioning technique, and communication. Summary statement feedback does the same for how they pull a case together. Both are low-stakes practice: learners see what they did well and what to work on next, try again, and get used to feedback long before the bedside. We treat what a VPE produces as feedback for the learner, not a verdict, so that routine feedback becomes ordinary and faculty can spend their time on the feedback only a person can give.

 

References

1 Al-Mously N, Nabil NM, Al-Babtain SA, Fouad Abbas MA. Undergraduate medical students' perceptions on the quality of feedback received during clinical rotations. Med Teach. 2014;36 Suppl 1:S17-S23. doi:10.3109/0142159X.2014.886009

2 Abdulnour RE, Gin B, Boscardin CK. Educational strategies for clinical supervision of artificial intelligence use. N Engl J Med. 2025;393(8):786-797. doi:10.1056/NEJMra2503232

3 Soderstrom NC, Bjork RA. Learning versus performance: an integrative review. Perspect Psychol Sci. 2015;10(2):176-199. doi: 10.1177/1745691615569000

4 Shute VJ. Focus on formative feedback. Rev Educ Res. 2008;78(1):153-189. doi:10.3102/0034654307313795