Imagine an evolved relationship between educator and student — one where they co-design the learning. Together, they find a problem worth solving. Then they dig into it.
Maybe even the fourth or fifth as the educator-learner relationship evolves to co-design learning.
There's a sentence I keep hearing, and I heard it again this week.
AI makes learning easier.
I was in dialog with researchers at ASU's Mary Lou Fulton Teachers College; people who have spent careers on learner variability, on the fact that no two learners arrive the same way. The sentence came up the way it always does: as a given. Something everyone nods at before moving on.
I've stopped being able to move on from it. Because easier is doing an enormous amount of work in that sentence, and nobody ever says easier at what.
Easier to get an answer? Easier to finish the assignment and close the laptop? Or easier for every learner — not the lucky ones — to get coaching that scaffolds them through the work that durable learning actually requires?
Those are not variations on the same claim. They are opposite ends of a spectrum. At one end, AI is a shortcut that quietly erodes learning. At the other, it is the first thing in the history of education that could genuinely personalize at scale.
And we now have evidence from both ends.
The shortcut end
A large-scale study of secondary students in China, reported on VoxEU, followed what happened when students used AI the way most students will use it if nobody intervenes.
Homework gains. Scores rose 18%, and time spent dropped 30%; from 64 minutes to 45.
Exam penalties. Closed-book monthly exam scores fell 20% within six months of adopting AI.
High-stakes exams. Entrance exam scores fell 18–24%, with the worst penalties after about two years of heavy reliance.
Read that sequence again, because the shape of it is the whole story. The homework got better. The learning got worse. And the gap took two years to fully show up — which means by the time anyone noticed, two years of learning had already been traded away.
The scaffolding end
Then there's a Harvard study in Nature's Scientific Reports, which built an AI tutor on purpose: grounded in learning science, designed to preserve productive struggle, adaptive in its scaffolding. It withheld. It made students do the work.
Students using it outperformed students in an active learning classroom taught by some of the best professors in the field.
Learning gains. Median post-test 4.5 versus 3.5, against a pre-test median of 2.75.
Time on task. 49 minutes with the tutor versus 60 in the classroom.
Engagement and motivation. Higher on both; 4.1 versus 3.6, and 3.4 versus 3.1.
Same technology. Opposite outcomes. The variable wasn't the model — it was whether anyone had decided what the thing was for.
For a moment, let's assume the hard part is solved
Let me grant the optimistic case for a moment, because that's where the interesting question lives.
Assume we could coach students on the appropriate use of AI. Assume we could build the guardrails and the controls that make responsible use the default rather than the exception.
Then every learner would have a personal learning coach in their pocket. Not a tutor — a coach. One that holds the line between support and effort. That explains the same idea six different ways until one lands. That goes deep when depth is what's needed and broad when it isn't. That catches a misconception in the moment it forms, and closes a gap the learner didn't know they had.
That is not a small thing. That is the thing education has wanted for a hundred years and never been able to afford.
And it raises a question I don't think we're asking enough.
Not what does AI do to learning. We're all asking that one.
What does AI do to the relationship?
The third seat
In a panel discussion at the convening, one of the researchers put something on the table that I haven't been able to put down.
He invited us to imagine an evolved relationship between educator and student — one where they co-design the learning. Together, they find a problem worth solving. Then they dig into it.
My first imagination of co-design - the educator brings what years of training actually buy you: the ability to hold space for a learner to go deep into something hard. To sit with the questions. To follow a thread of inquiry without rushing it to a conclusion. To stay steady through the emotional roller coaster that any real struggle produces — because struggle is the point, and struggle feels bad, and someone has to be there who knows the difference.
The student brings the range. Free to go find the knowledge, the tools, the resources with AI as the thing that makes the going possible. Down the rabbit holes. Into the blind alleys. Back out again.
Partners in the journey. The educator as the keeper of emotional and mindset space, the learner as the intrepid one.
What strikes me about this is which half AI can't do.
AI can explain. It can scaffold, adapt, assess, and never get tired. What it cannot do is be the person who noticed you were about to quit and decided not to let you. It cannot hold a relationship you don't want to disappoint. The thing that gets a fifteen-year-old through the part where it stops being fun is almost never information.
So maybe the coach in the pocket doesn't replace a seat at the table. Maybe it takes a third one. The educator stops being the keeper of knowledge — a job they were never going to win against a machine — and becomes the keeper of the conditions under which hard things get finished.
That's not a diminished role. I'd argue it's the harder one, and the one we've been underpaying for decades.
To be continued …
While I am still digging into what co-design looks like, what I'm fairly sure of is this: the two studies above aren't in conflict, and treating them as a debate about whether AI is good for learning misses what they're actually telling us. They're telling us the technology has no opinion. We do.
The question isn't whether every student will have a learning coach in their pocket. They will.
The question is what the adults in the room decide they're for.
The exploration continues.