Nobody has handed us the rule book, so we’re writing ourselves.

*Banner image was created using Gemini NanoBanana 2 (Flash Image 3).

Last year, FWOC (first week of class) felt like the Cold War — mutually assured anxiety, as I will frame it. You pull up the syllabus and scroll down to the academic integrity section under policy. Some professors outright banned AI usage; others said nothing at all, but the silence felt louder than the spoken rules.

This year, the same tension persisted but certainly with a different taste. The chaos of "Can I use it? Is it beneficial?" settled into something more deliberate — still an uncertainty, but a more structured one that feels less like negligence and more like a choice. We're sitting in a transitional buffer, held open for us to actually figure out what these tools mean before anyone commits it to paper.

I. Intent-centric, CS Perspective

As a CS major, I can recall how, a year ago, the campus divided up clearly into two camps: one saying "don't use it while you're learning" and the other saying "it's inevitable, so be strategic." Both were right, and neither gave you much to work with on a Tuesday night when you had a problem set due.

What's changed is that the conversation has moved past whether to how, and the how has gotten surprisingly specific. Professors are now explicitly allowing LLM tools in assignments but docking points for dirty code, for inefficient logic, for output that was clearly generated and never given a second look. Autograders are preliminary measures and manual adjustments yield the final grade. They are asking whether you understood what you submitted and why you made each line of choices.

I've talked to people in my major who genuinely miss the old grind. Not masochistically, but because something real happened in the hours you spent wrestling with errors before submission. There was a human intuition being built, a problem-solving muscle that formed in the friction. And they are right about certain things. AI is not just hallucinating with its responses but with our conception of our own fluency and competency with the language and the tools. Classrooms are seeing a split between "real coders" and "vibe coders," but that distinction alone cannot capture or represent the quality of the output. "What does it mean to actually know how to code?" is perhaps the question most of us want to see answered, and one we're still struggling with as we navigate.

---

II. The Gap Between Policy and Practice

A 2026 Gallup State of Higher Education study found that over 60% of college students are using AI tools daily, yet roughly 40% of institutions still officially discourage or restrict their use. Meanwhile, a 2026 Digital Education Council survey found that while 92% of students are engaged with AI in some capacity, only 30% feel that the guidelines around it are actually clear.

These numbers make clear the confusion all of us have experienced at some point. Most of us are using these tools, yet most of us are doing it in the fog. But I've come to think the fog is at least partially intentional. I was talking to a faculty member who framed it this way: by not issuing a blanket policy, the university is effectively granting each faculty member the agency to discover what AI means for their domain — what it breaks, what it enhances, what it makes irrelevant — without being forced into a framework that doesn't fit. Surely, a one-size-fits-all AI mandate released at the beginning of the semester would have become an embarrassment by the time we reached finals. Fortunately, we had room. Uncomfortable, poorly-lit room, but still room.

That doesn't lessen the confusion when you're a student trying to figure out whether citing ChatGPT is required or whether your professor even wants to know. But it reframes the frustration. The gap in policy is there by purpose, to let actual practice catch up to any rule that might try to govern it.

---

III. Humanity…in the Process

I've collected from non-CS friends how the trajectory has been for them over the past year. According to them, a year ago, "AI in the humanities and social sciences mostly meant a slightly better Google." They characterized it as useful for summaries, guilty as a shortcut, and generally mistrusted — I think the mistrust was earned, since the tools were less reliable and the use cases were less obvious. Skepticism persists, but the degree of it has definitely softened because the models got better and people got used to them. AI-backed summarizing and primary research tools have evolved both in usability and credibility and have been integrated into a lot of workflows.

What's more interesting to me is what's happened in certain classrooms that, a year ago, seemed like the last places AI would show up. Some arts classes are now asking students to create something and then place it next to what an AI produced from the same prompt — not to judge the AI, but to interrogate yourself. Where did you make choices the model didn't? Where did the model go somewhere you wouldn't have thought to go? Some writing classrooms are doing versions of the same thing. It sounds gimmicky until you're actually sitting with both pieces and realizing that the comparison is teaching you something about your own voice. These are some of the most interesting pedagogical moves I've seen this year, because they use AI not as a tool for efficiency — like people like I do — but as a mirror. If the intentional gap in policy exists so that faculty can discover how humans and machines naturally diverge in their domain, then these classrooms are where the real answers are being written.

---

Outro

Here's where I keep landing: the transition we're living through is about a shift in what the student's job actually is. For a long time — in CS especially, but really everywhere — the job was to be a producer. Write the code, write the essay, make the thing. The craft was in the making.

What's emerging now is something more like curation. The job is increasingly to have taste, judgment, and intent — to know what you want well enough to recognize when you're getting it and when you're not. That's a different skill. It's not harder or easier, exactly, but it requires a kind of self-knowledge that the old model of education didn't really ask you to develop explicitly. You had to know your field. Now you have to know yourself in relation to it.

The gap in policy has to close eventually. Someone will write the rulebook, and it will be outdated, and then someone will revise it, and it will be outdated again, and that's just how institutions work. But right now, in this particular moment, the how is being written by the students sitting in office hours asking whether they can use Copilot, and the professors staying up late figuring out what it means that their students can generate a passable first draft in eleven seconds, and the writing classrooms putting two versions of the same essay next to each other and asking: which one is yours, and how do you know?

---

Written in Spring 2026 by a Duke undergrad who has thought about this way too much and will probably think about it more.