What Happens When You Stop Practicing the Hard Parts

The finding from researchers who study writing and cognition is not subtle: students using AI writing tools are producing more polished work while retaining less of the underlying reasoning that the writing was supposed to build. Polished outputs, flat comprehension. The research language is careful; the classroom observation is not.

Jenna Rourke teaches junior English at a public high school in Boise — eleven years in, by any measure an experienced reader of student work. Last spring she assigned a personal essay, the kind where the prompt is deliberately open and the point is to make something that couldn't come from anyone else. She got thirty-two back. "Smooth," she told me. "They were all smooth."

When she asked students to talk through their essays in class — explain a sentence choice, say why they'd started where they did — several struggled. Not all. But enough that it caught her attention. The thinking that should have preceded the writing hadn't always happened.

The finding is counterintuitive only if you think of essays as outputs. Think of them as exercises — a technology for organizing thought — and it makes sense. You don't get stronger by watching someone else lift.

The analogy feels almost too simple, but it holds. Writing is one of the few activities where the difficulty is the benefit. The struggle to put an idea into a sentence — to figure out what you actually think, to find the word that isn't quite right and then find the one that is, to sit with a paragraph that almost works and keep pressing until it does — builds something that has no shortcut. Cognitive scientists call it desirable difficulty. The resistance is the point. Remove it and the output improves while the underlying capacity stays flat or declines.

What Writing Was For

High school and college writing assignments have always been imperfect. The five-paragraph essay is a blunt instrument. The research paper often produces cut-and-paste work that no one, including the student, is proud of. These are real limitations, and teachers know them.

But even imperfect writing practice builds something that AI-assisted writing practice may not: the experience of being stuck, of not knowing what you think until you've tried to write it, of discovering in the middle of a paragraph that your argument doesn't hold. These are generative failures. They produce knowledge about your own thinking. What researchers who study skill acquisition call the slow process of learning by doing — making a bad first draft and understanding, from the inside, why it's bad — is not a pleasant process. It is also not optional if the goal is to actually learn to think.

That's the thing that's hard to explain to students who are being rational about their immediate situation: the grade matters, the deadline is real, the AI is faster. All of that is true. What's also true is that skipping the hard part of writing doesn't just affect the quality of this essay. It affects the quality of the thinking that comes after.

The Competence Gap

Jenna's smoothness problem has a name in assessment circles: competence inflation. It's been a concern in education long before AI — grade inflation, credential creep, standardized test prep that produces scores without producing knowledge. AI writing assistance is a significant intensification of a pre-existing problem, not a new problem from scratch.

What makes this intensification different is the scale and the seamlessness. Copying someone else's essay leaves traces. Asking a friend to write it produces something stylistically foreign. AI assistance is invisible and infinitely available. A student who uses it on every assignment throughout high school will have produced thousands of pages they didn't actually write — and will have missed thousands of hours of a practice that builds something harder to measure than a grade.

I almost dropped this section of the argument because it felt like the kind of thing adults have always said about whatever shortcut the current generation found. But Jenna's description of the oral defense problem stopped me. She wasn't describing students who wrote bad essays. She was describing students who wrote good essays they couldn't explain. That's a different thing entirely.

Teachers are trying to adapt. Hand-written in-class essays. Oral defenses. Process portfolios that show drafts and revisions. Jenna now asks her students to submit not just the final essay but a record of their choices — a paragraph explaining why they started where they did, what they cut, what changed. She's looking for evidence of a mind at work. Sometimes she finds it. Sometimes the record is as smooth as the essay.

What Gets Kept

The answer is not to ban AI from classrooms. That ignores the reality of what students face outside school, and enforcement would fail anyway. The answer is to be honest about what writing is for — not as a way to produce text, but as a technology for developing a mind — and to design assignments that can't be evaded by outsourcing the difficulty.

That means more oral work, more visible process, more constraints that force the thinking to happen in front of you. You can use a calculator and still not understand arithmetic. You can use AI to write an essay and still not know how to think through a hard problem in prose.

Jenna told me she sometimes asks her students, at the end of a class discussion, to write one sentence — handwritten, on paper — about what they actually think. Not a thesis, not an argument. Just what they think. Some of them find it surprisingly hard.

She told me this with more equanimity than I'd have. What I kept thinking, driving home, was that we've been through versions of this argument before — about calculators, about spell-check, about the internet — and the pattern is always that the new tool wins, the skill that the tool replaces atrophies in the general population, and we argue for twenty years about whether that matters. It usually matters. We usually decide it doesn't matter enough to change anything.

I don't know how this one ends. The tool is better than the last ones, and the skill it's replacing is harder to measure than arithmetic. That's a bad combination. But I also don't think Jenna is wrong that you can design around it — she's doing it, in one classroom in Boise, and it's working for some students. Whether that scales is the question nobody at the policy level is seriously trying to answer yet.