I learned medicine in hallways, in the classroom, and outside.

I learned how to apologize to a patient by watching my attending pull up a chair and say, “I should have done better.” I learned how teams actually work by overhearing tense but respectful disagreements during sign-out. I learned what it meant to belong by being given responsibility before I felt ready. None of those lessons happened on Zoom or Teams.

In the post-pandemic era, many medical schools have retained hybrid and virtual formats not just for lectures, but for advising, mentoring, professionalism sessions, and identity-shaping conversations. Flexibility and access are the usual justifications. Now artificial intelligence (and increasingly agentic AI) are accelerating this shift. What began as a temporary adaptation risks becoming the default infrastructure of medical education. That should give us pause.

The limits of virtual identity formation

Medical education is not just about transferring information. It is about practice socialization and identity formation. Professional identity is not something students download or ask GenAI. It does not emerge when they log in, or when a GenAI tutor answers their questions instantly. Identity forms when students are invited into live and real learning clinical communities and entrusted with authentic responsibility.

I recently spoke with a student who had met with an AI-powered advising tool before seeing a faculty mentor. The AI efficiently mapped specialties to interests, predicted competitiveness, and suggested a career trajectory. When the student later met a clinician mentor in person, the conversation shifted. They talked about doubt, meaning, and the emotional cost of certain paths. The AI was accurate without meaning making, risk, and safety for that student. The human was formative.

I have watched first-year students attend virtual professionalism sessions with cameras off and microphones muted. The content is delivered efficiently. The session ends on time. Nothing goes wrong, but nothing really happens either. AI now promises to make these experiences even more scalable: chatbot mentors, automated reflections, simulated ethical dilemmas. Yet none of these replace the moment when a learner watches a physician struggle, reflect, and choose wisely in real time with a patient at the bedside or in a classroom.

Contrast that with a student sitting in a cramped conference room after rounds. A resident vents appropriately about a difficult family meeting. An attending reframes the frustration without dismissing it. The student watches clinicians wrestle with uncertainty, emotion, and responsibility. That is professional identity formation. And it does not scale.

Skill development in the real world

During the pandemic, students adapted admirably to virtual learning. But educators sensed what evidence later confirmed: fewer informal interactions, disrupted socialization, and a growing sense that students were learning medicine without fully becoming physicians. AI risks deepening this trend by making it easier to substitute simulated competence for lived participation. When identity formation becomes something students must self-assemble with AI as a stand-in for mentorship, the burden quietly shifts from institutions to learners.

The same risk applies to skill development. Much medical knowledge can be delivered online. AI can summarize, quiz, coach, and even simulate patients. But medicine is practiced in complex social environments, not controlled interfaces. Take history taking. On paper, it is a checklist. In practice, it is choreography: when to interrupt, when to sit with silence, how to read discomfort, how to earn trust in five minutes. I increasingly see students who sound polished and precise yet struggle at the bedside not because they lack knowledge, but because they have had fewer opportunities for supervised, embodied practice with real patients under real constraints. Hybrid education is efficient and spontaneous. AI promises personalization without authenticity. Together, they risk crowding out the messy, inefficient, high-yield experiences that turn knowledge into judgment and wisdom.

The burden on faculty

Faculty feel this pressure acutely. Hybrid systems already increase workload through duplicated content, platform management, and constant availability. Agentic AI adds new expectations: reviewing AI-generated feedback, overseeing automated assessments, and responding to institutional pressure to “do more with less.” While AI is marketed as a faculty extender, it often consumes the very resource students value most: caring by faculty.

When educators spend more time managing systems than observing learners, the apprenticeship model erodes. The AI also accelerates a quieter shift: the normalization of “default virtual.” When advising, feedback, remediation, and professionalism can all be mediated by intelligent systems, it becomes increasingly difficult to justify time-intensive, in-person alternatives. Convenience begins to masquerade as progress and innovation. This is not an argument against AI. It is an argument against uncritical adoption.

What we must protect

Medical schools must be explicit about what should not be automated or virtualized: clinical apprenticeship, mentoring, small-group coaching, and team-based work. These are not nostalgic holdovers. They are the core technologies of professional formation. The question is not whether AI can help educate future physicians. It can. The question is whether, in our pursuit of efficiency and scalability, we are willing to protect the experiences that teach students not just how to act, but who to become. If we are not careful, we may optimize medical education for convenience and intelligence while quietly disinvesting in identity formation and development.

Vijay Rajput is an internal medicine physician.