Picture a pretty ordinary assignment. A student is supposed to read something, decide what matters, make a claim, and explain the reasoning behind it. The student opens an AI tool, and within a few seconds it can suggest the claim, choose the evidence, organize the answer, and produce a polished paragraph. The finished work may look terrific. But what, exactly, did the student learn?
That is why I don't think the most useful question is, "Should students be allowed to use AI?" That question is so broad that it almost immediately turns into an argument about whether AI is good or bad. The more practical question, the one I keep coming back to, is this: What do I actually need the student to think through for themselves?
Help isn't the problem
After 26 years in the classroom, I certainly don't believe students should have to do everything without help. They have always used notes, examples, calculators, conferences, vocabulary support, extra time, and plenty of other assistance. Good teaching has never meant standing back and watching students remain confused.
The question is whether the help gets students into the work or quietly does the work for them. If a student doesn't understand the directions, define them. If an unfamiliar word is blocking access to the whole assignment, explain it. If one more example would help a student see how to begin, provide the example. There is no great educational virtue in preserving difficulties that have nothing to do with what we are trying to teach.
But suppose the assignment is asking students to decide which evidence matters, develop a claim, work through a calculation, explain their reasoning, or choose how to revise a piece of writing. Those are not incidental obstacles. Those are the things we wanted students to learn how to do.
Protect the hard part that matters
AI can be helpful without taking over those decisions. It might ask a student a good question, point out a step that has been skipped, clarify a confusing direction, or remind the student of a strategy they have used before. The student still has to figure out what the question means and decide what to do next.
The line gets crossed when the tool begins making the consequential choices in the background. If the student is supposed to choose the evidence, the tool should not choose it. If the student is supposed to make the argument, the tool should not make it. If the student is supposed to decide how to revise, the tool should not simply rewrite the paragraph and hand it back.
Otherwise, we can end up in a strange place. The assignment looks better, the writing sounds smoother, and the answer may even be correct, but the student has had less reason to think. We have improved the product while weakening the process that was supposed to produce the learning.
Draw the line before students begin
I think students deserve to know where that line is before they start. This does not require a five-page AI policy filled with language no student will read. It can be much simpler: Here is the kind of help you may use. Here is the thinking you still need to do. Here is what you should be able to explain when you are finished.
If we cannot explain that boundary clearly to students, we probably have not figured it out clearly ourselves. Before the next assignment, I would put a few questions at the top of my planning notes:
- What am I actually trying to help students learn?
- Which decision or mental move do I need them to practice?
- What kind of help would remove a needless barrier?
- What kind of help would give away the thinking?
- What should students be able to explain in their own words?
- When will they try something similar again with less help?
That last question may be the most important. A student can produce an impressive answer today without being able to do anything similar tomorrow. Supported performance and actual learning are not necessarily the same thing, especially when the support is powerful enough to produce most of the finished work.
Come back to the learning
At some point, students need another chance to retrieve the idea, explain it, or use it in a different situation. Can they recognize the same kind of problem later? Can they explain why they chose a particular piece of evidence? Can they find the missing step in their own reasoning? Most importantly, can they do more with less help?
I would rather hear a student say, "I chose this evidence because it supports my claim," than receive a smoother paragraph the student cannot explain. I would rather hear, "I changed this part because I realized my reasoning skipped a step," than see a flawless revision that somebody or something else effectively made for them. The goal is not to catch students using a tool. It is to keep students responsible for the choices the assignment was designed to help them practice.
Why I'm building Coach
This is a big part of the problem I'm trying to work on with Coach. I want students to be able to get useful help without simply handing their thinking over to AI. I want teachers to decide what kind of support is appropriate for a particular assignment, and I want the work students do along the way to lead to better teacher questions, not automated judgments about the student.
I don't pretend that Coach has solved all of this. Nobody has. But I do think we can begin in a sensible place: choose one assignment, identify the thinking that matters most, remove the barriers that do not matter, and make sure the student still has to do the part that does.
Then we come back later and see what the student can do. That, after all, is the part that matters.