When AI Can Do More of the Thinking, School Has to Change

Every teacher knows that a student who accomplishes more isn’t necessarily becoming more capable. Peers, parents, tutors, and tools have always propped up student performance. AI is different in degree, and the difference matters. A student working with AI can produce work far beyond what she could do alone, and used well, AI is able to extend what she can investigate, create, and understand. But the human-plus-AI system can grow more capable while the human inside it develops less. From the finished work alone, we can’t necessarily tell which is happening, and that should change how we think about school.

AI is also changing the world students will enter. They will live and work in a world where cognitive work—interpreting information, generating options, solving problems, making recommendations—is easy to delegate. Schools must become far more deliberate about developing students’ capacity to direct their own thinking, judge what is worth pursuing, and take ownership of their intellectual lives. Teaching students to use AI well is part of that, but only part. It will take redesigning curriculum, assessment, and classroom practice so that students develop more agency than schools have ever needed to intentionally cultivate in the past.

Agency is more than choice

Agency has become a familiar word in education, and we often use it too casually. We tend to equate it with choice: selecting a book or pursuing a topic of interest.

Choice can help develop agency, but it isn’t enough. A student can choose to have AI write the essay. A student can choose the easiest route through an assignment. Autonomy can be exercised without ever asking whether the choice is worthwhile.

I think of agency in an AI-shaped world more expansively:

Agency means knowing what I want, understanding why I want it, examining whether it is worth wanting, and retaining ownership of the judgments that guide my actions.

My definition of agency understands that now and in the future students will work with systems that suggest what to read, what to write, which questions to ask, and what comes next. The question for us (and them) is whether they learn to notice those influences and govern them rather than simply follow them.

Of course, agency is not the whole of what schools develop. Students also need knowledge, judgment, creativity, relationships, and character. But agency shapes how they use those capacities, toward what ends, and with what responsibility.

Nor is agency a solitary achievement. Students learn what is worth wanting, and learn to judge well, in the company of others: arguing with peers, being questioned by teachers, answering to a community to which they are responsible. A mature student decides what is worth doing, understands why, chooses the people and tools that can help, and accepts responsibility for the judgments that follow. AI makes that developmental work more urgent.

Outsourcing thinking has become easy

There is nothing inherently wrong with outsourcing cognitive work. Adults do it constantly: we use calculators, consult experts, and collaborate with people who know what we don’t. AI adds a remarkably capable new participant. The question is what happens to the student when some of the thinking is delegated.

Imagine two students using the same AI tool…

One has noticed something odd: her school moved to a later start time two years ago, and grades rose. She wants to know whether the changed time caused grades to improve. She asks AI for ways to investigate, and it points to research linking later start times to better outcomes. She questions whether that research settles her school’s case, rejects some suggested approaches, and follows one into unfamiliar territory: the school’s own grading and attendance records. In her investigation, she discovers the school also revised its grading policy that year, and thus her final conclusion differs from the one AI first proposed: the evidence can’t yet show whether the start time caused improved grades.

The other asks AI whether later start times improve grades, accepts its summary of the research, uses its evidence, and asks it to polish the result.

Both may submit impressive work. Only one of them directed it. And, as it turns out, only one of them is likely correct. And the finished artifact increasingly won’t tell us which. A strong essay or analysis may represent substantial human thinking, or very little.

That is why schools can’t solve the AI challenge mainly by getting better at detecting AI use. We need to get better at developing students who remain intellectually responsible when they use it.

Four shifts

If that is the goal, adding AI guidelines to existing practice will not be enough. I see four necessary shifts schools will need to make:

1. From completing assigned work to understanding purpose: school is full of purposes set by adults: read this chapter, write this paper, etc. There are good reasons for that. Young people need structure, and teachers have expertise students don’t yet have.

But students who spend most of their education pursuing someone else’s purposes get little practice setting and examining their own. AI raises the stakes. A student who can produce almost anything with AI needs practice with a prior question: What is worth producing?

That means more than occasionally letting students choose a topic. Students might develop several possible research questions before selecting one, then explain why it matters and what they hope to understand that they don’t yet. In a ninth-grade unit on the Great Depression, a student drafts five questions and uses AI to test them. She drops the ones AI can answer well in a paragraph, like what caused the Depression, and keeps one it can’t settle: why her town’s bank survived 1933 when two nearby banks failed. If AI can answer a question in a paragraph, it probably isn’t worth a month of her attention. AI can help generate possibilities. Students should learn to decide which deserve pursuit.

2. From producing answers to exercising judgment: schools have always paid close attention to whether students arrive at good answers. That still matters, but in an AI-rich world, answers are easy to get, good judgment is not. AI can generate plausible interpretations of a historical event, solutions to a problem, or competing explanations of a phenomenon, but students need to decide among them. That takes reasoning, context, evidence, tolerance for uncertainty, and knowledge. Knowledge still has to be built, often without AI, because no one can evaluate a claim in a domain they don’t understand.

So curriculum and assessment should demand more judgment. Ask students to compare alternatives, weigh evidence, identify assumptions, and defend conclusions. Ask what evidence would change their minds, and what an apparently strong argument leaves out. Ask not only “What is the answer?” but “Why should we believe this?”

In chemistry, a lab group’s yield comes out 15 percent below prediction. AI offers three explanations: an incomplete reaction, measurement error, or product lost during filtering. The students rank the explanations using their own lab notes and design the one test that would tell them apart. A plausible answer is often where judgment begins.

3. From evaluating finished products to making thinking visible: as AI becomes embedded in ordinary intellectual work, schools need greater visibility into how students think. I don’t mean surveillance, or treating every assignment as a cheating investigation. I mean giving more educational value to the process itself. Ask students to explain their consequential decisions. Why did you pursue this question? Why did you reject that source? Where did AI change your thinking? Where did you disagree with it? What judgment in this work is distinctly yours? A history teacher asks students to attach a half-page decision log to their research papers. It names three moments where AI suggested something, what the student did, and why. The log then anchors a five-minute conference, which often reveals more than the paper itself.

A short conversation between teacher and student may reveal more about learning than another page of finished prose. Drafts, annotations, oral defenses, and a student’s reconstruction of the reasoning behind a conclusion can do the same. Assessment should value this evidence of thinking alongside the finished artifact. That helps with academic integrity, but it matters more for learning: students get better at directing their thinking when someone who takes it seriously asks them to examine it.

4. From teacher-directed environments to progressively greater intellectual responsibility: we shouldn’t expect a ten-year-old and an eighteen-year-old to work with AI and exercise the same intellectual independence. Schools need a deliberate progression.

In the younger grades, AI should be largely kept away from student use, letting adults shape student learning. They set boundaries, select resources, model judgment, and decide when tools are appropriate. A fourth-grade teacher uses AI in front of the class, thinking aloud: “That sounds confident. How could we check it?”

Around the start of middle school, students can begin using AI directly, within structures their teachers design. Seventh graders might ask AI for help choosing a persuasive-essay position, phrasing the request three different ways. They compare how each phrasing steered the answer before deciding what they actually think. Students learn to notice how their learning environments shape them: how incentives drive behavior, how peers sway judgment, how technologies direct attention, and how an AI system can set a task’s direction before the student realizes it.

By upper school, students should be building and governing those environments themselves. A student might draft her own AI-use plan for a major project: where AI will help, where it would get in the way of what she needs to learn, where another person would be more useful, and where she will work unassisted. She then defends that plan and accounts for it at the end.

The point is not to give older students progressively more freedom with AI, it is to expect progressively more responsibility from them. The goal is a graduate who can decide when, why, and how to use AI, and who understands that sometimes the right decision is to do the thinking without it.

What this asks of schools

These needed shifts will not be easy. Teachers are reluctant to change their practices and administrators are reluctant to require it. And teaching approaches like these—student conferences, oral defenses, and careful feedback on reasoning—take time. They also require teachers who exercise these judgments themselves and have room to know their students’ thinking well.

Independent schools are unusually well positioned for this work. Small classes, advisory programs, and sustained teacher-student relationships are the conditions it requires. We usually describe those strengths in terms of care and belonging, but in an AI-shaped world, they are also the infrastructure for developing judgment.

A different kind of preparation

AI will keep getting better, and schools should take advantage of it. But access to greater capability is not the same as developing greater human capacity. And because students will enter a society where powerful systems participate in work, decision-making, and civic life, the stakes reach well beyond academic performance.

Return to my earlier example of the two students investigating school start time and grade distributions. They had the same AI tool. What separated them was a set of habits: deciding what was worth pursuing, questioning what they were given, and keeping hold of their own judgment. Those habits can be taught. The work ahead is building schools that reliably produce the student maintained agency.

When AI can do more and more of the thinking, how must school change so that students become better able to think, and direct their thinking toward what is worth doing? That is the question curriculum committees and academic leadership teams must be prepared to begin to address this year.

How I wrote this:

I develop the ideas for my posts on my own, with my human brain, and I craft a thesis, key points, and prose for every post all by myself. I use various Al tools to review and revise some of my prose and to test certain ideas, much like a human editor might, yet the content of each post comes from my brain. Like I’m recommending schools ask of their students, I use a writing process that preserves my thinking, my agency, while using AI to improve the output.

Mike Gwaltney is Vice President of Consulting and Learning, and a Senior Consultant at Independent School Management (ISM). He is a long-time independent school educator and former Head of School. He has been at the forefront of academic technology since the early 2000s and is widely considered the “OG” of blended and online learning in independent schools. He holds a Masters in Educational Technology, and writes about student agency, teaching and learning, and leadership and governance of AI in independent schools.