Harvard’s Student Voice: What Undergrads Want Faculty to Know About Using AI

T. Alexander Puutio, a professor of economics at Harvard, observed an odd phenomenon with his students’ essays in the fall of 2022. Suddenly, writers he knew to be strong B+ students were turning in work that was polished in ways that didn’t resemble anything they had previously produced, and it was strewn with Oxford commas and em dashes. The writing was skillful. The voice had vanished. He quickly identified it as the specific blankness of text that has been processed rather than written, which has since been dubbed “AI slop” with a high degree of accuracy.

Puutio took a different approach than outlawing the instruments that made it. He now mandates that students use AI in all of their assignments. This decision, which he described in an essay for Business Insider in March 2026, is based on a set of guidelines intended to do something that most university AI policies aren’t yet capable of: differentiate between AI that does the thinking and AI that supports the thinking.

The distinction seems straightforward. In reality, it is the whole issue. Undergraduate students at Harvard are as aware of this as their professors are, and in some ways, they have been more openly dealing with its complexity. Developed through interviews with seven faculty members and 27 HGSE graduate students, a guide released by the Harvard Graduate School of Education in January 2025 paints a picture of students who are considering AI in ways that aren’t always reflected in institution-wide policy documents. These students aren’t attempting to get away with anything. They are students who have already incorporated AI into their work processes and are now, quite understandably, requesting that their teachers catch up.

Even though it seems hard to deliver, what they want is not complicated. Instead of the ambiguous language about “responsible experimentation,” which Harvard’s general AI policy currently offers, they want explicit policies that specify exactly what constitutes authorized use in each course. Instead, they want actual guidance from the instructor who is grading their work. Professor Karen Brennan, who co-wrote the guide and oversees the Creative Computing Lab at HGSE, compares the fear surrounding AI to the “moral panic” that followed the development of the pocket calculator. This fear proved to be partially justified, partially misguided, and primarily resolvable through careful integration rather than outright prohibition. Similar statements were made by the students in her study. When instructors explained why they were assigning a certain task, they said it was actually helpful because it made it easier for them to decide what to give to AI and what to keep for themselves.

The HGSE guide quotes a student who said, “Really think about what you want at this moment,” which seems to have more wisdom than most institutional policy statements. Do you want to learn or just finish the task at hand? Perhaps that question ought to be at the top of every American course syllabus at the moment. The students who are inquiring are aware of the limitations of AI. They cautioned one another about the challenging learning curve, the need for several rounds of quick refinement to produce meaningful results, and the moment one student, six hours into a session that should have taken two, realized that sometimes using your own skills is just more efficient than battling the machine.

One example of this in action is the framework Puutio employs in his own classroom. AI for research and synthesis, AI as an editor and critic once the argument has been developed, but never AI creating the argument itself. The student must be the one who thinks. When that line is clearly drawn, the assignment’s nature is altered without giving up on technology. It’s difficult to ignore how different that is from the binary decisions that the majority of universities continue to make: either completely prohibit it or say nothing helpful and hope for the best.

Fundamentally, what Harvard students are requesting is honesty. They are aware of the existence of AI. Regardless of policy, the majority of them use it in every course. They want academics who are prepared to confront this reality head-on, not act as though it will disappear or treat every discussion of AI as a disciplinary issue, but rather sit with the true complexity of a technology that, depending on how it is applied, can either enhance or detract from learning. This is already being figured out by the students. Whether their institutions will advance fast enough to be beneficial is the question.