The Semester I Made Every Student Use AI: A Field Report

After sixteen weeks of requiring AI in every assignment, the most surprising thing I learned is that the debate raging in faculty lounges misses the point entirely. Critics warned me students would stop thinking. Evangelists promised they'd finally start. What actually happened was messier and more interesting: some students wrote better papers than ever before, some produced sophisticated-looking work that revealed shockingly shallow reasoning when I asked them to defend it, and a few—okay, more than a few—spent the first four weeks trying to figure out how to trick me before accepting that I genuinely wanted them using the tools. The writing quality didn't collapse or soar. It stratified. Students who already had strong metacognitive skills leveraged AI to do things they'd never attempted before. Students who were coasting got faster at producing surface-level work. The technology didn't change my students. It revealed who they already were. What changed most dramatically wasn't the output—it was engagement with the revision process. Here's what caught me off guard: students who previously submitted first-draft thinking began treating AI-generated text as a first draft they needed to wrestle with, not unlike how they might respond to a peer's rough draft. They argued with the AI's conclusions. They pushed back on its examples. For the first time in years, I saw students genuinely confused about what was 'their idea' versus what the tool had suggested—and that confusion, painful as it was for some of them, opened up conversations about intellectual ownership that I'd never been able to catalyze before. The engagement wasn't universal, but for the students who leaned into the discomfort, it was transformative. The honest accounting: I saved three assignment types that I almost killed. The analytical essay, which benefited from AI helping students see patterns they'd missed in their own reasoning. The peer review exercise, where students used AI to generate feedback then had to evaluate and revise that feedback—turning a often-superficial task into something requiring genuine judgment. And the literature review, where AI handled the organizational scaffolding while students made the interpretive calls that actually mattered. But I also discovered that two assignments I'd considered core to the course were doing almost nothing I assumed. The weekly reflection journals had become performative compliance exercises long before AI existed—I just hadn't noticed. And the thesis statement workshop, it turned out, was teaching students to write for me rather than think through ideas. AI didn't break these assignments. It exposed that they'd already been broken. If you're considering running this experiment, here's my practical advice: design for the uncomfortable middle, not the extreme outcomes. Plan specific checkpoints where students have to articulate what the AI did versus what they contributed. Accept that the first month will be noisy as students test your actual intentions. And prepare yourself to discover that some of your most sacred assignments were never doing what you thought they were. That's not a failure of AI. That's the real value of the experiment—it forces you to examine your pedagogy with fresh eyes. The tools will keep changing. What you learn about your own teaching in the process might be the only thing that stays useful.