In late September, on a working call, one of the sharpest technical minds I know said something I haven't stopped thinking about.

Aman Sharma is the cofounder and CTO of Lamatic.ai, an agentic AI development platform. He chairs the AI Collective's Technology Committee, is its Florida director and guest-lectures at universities. He builds the kind of systems most of us are still learning to use. And he said, calmly:

"In the next 10 years, I predict that no one would be teaching to humans, and nobody would be learning from humans."

He called it "a very strong opinion" and added, "I could be completely wrong here." In his model, teachers teach the AI the essence of what they know (he compared it to a zip folder), and the AI unpacks it for each learner, one-on-one. He was clear about who worries him most: students in smaller cities, the people he thinks the AI gap "is going to hit… the hardest."

I agreed with him completely, and it shocked me. I love teaching. At the University of Toronto, I built a three-course certificate in multimedia storytelling and taught it. Now I was agreeing that in ten years no one would be doing that work.

Then I kept hearing it. Eleven days later, the same idea came up three times in one day: in a room of educators, over lunch with two business owners who thought it might happen, and on a call with a newly retired colleague who put it plainly: "There's gonna be a transition."

So I've stopped asking whether it will happen. I'm asking what a learner will still need from a person when it does.

What does the teacher become?

People have been trying to build a machine that teaches for about a hundred years. In the 1920s, Sidney Pressey built one that could ask a question and check the answer. B.F. Skinner built his own versions later, and the Smithsonian keeps a collection of these teaching machines. They could quiz you. They couldn't explain an idea, put it another way or answer your follow-up question. Today's AI can do all three.

For most of that century, the idea was ahead of the technology. That has changed, and there's now research worth paying attention to. In a randomized trial at Harvard, physics students who used an AI tutor designed for their lessons learned significantly more, in less time, than students in an active-learning class on the same material. They also said they felt more engaged and motivated. That's one carefully built tutor, used for specific lessons, not a whole course. It's a narrow result, but a real one.

In Edo State, Nigeria, a six-week after-school program that used AI tutoring improved students' English scores. The World Bank researchers are careful about why. Teachers introduced the topics and helped students as they worked with the chatbot, and the sessions followed the curriculum. The researchers credit the whole program and say they can't separate out the chatbot's share.

Look at the people around the software, though. Someone designed the Harvard tutor. In Nigeria, someone matched the sessions to the curriculum, and teachers worked alongside the students. Taking the teacher out of one conversation isn't the same as taking people out of learning.

Alpha School has built its whole day around this idea. According to the school, students spend two hours on academic apps, and the rest of the day goes to projects, workshops and life skills. The adults aren't called teachers. They're called guides, and the school says their work includes mentorship, motivation and emotional development.

That's Alpha's own account, not an independent evaluation of its results. But it isn't a school without adults. It has changed what the adults are for.

That matters well beyond one school. Spending less time delivering lessons could give teachers more time for the people in front of them. It could also become an excuse to cut people out. The software won't decide which. We will.

The word "guide" stayed with me. When I first tried to picture what a teacher would do in Aman's future, my mind went blank. I couldn't see the human doing anything at all. Then I had a flash of the Socratic exchange: the question that makes you think, instead of the answer that lets you stop. It was only later that I realized I'd been picturing something very like Alpha's guide.

So what does a guide actually do? Less teaching, for a start. Beyond that, I see three jobs, and none of them is new.

Asking more than telling. In a study of Tutor CoPilot, an AI system coached human tutors in real time while they worked with students. Tutors who had it asked more guiding questions and gave away fewer answers, and their students were four percentage points more likely to master a topic. The AI didn't replace the tutor. It made the tutor more of a questioner.

Making sure the learner does the work. Having a tutor doesn't mean anyone uses it. A two-year randomized trial in 18 Tennessee middle schools, reported in a 2026 working paper, gave students Khan Academy and its AI tutor, Khanmigo, during remedial math sessions. Their math scores improved. But the students made little real use of the tutor, and the gains looked like those from ordinary Khan Academy practice. The study couldn't separate out what the AI added.

Someone still has to notice when a student is avoiding the hard question, and help them stay with it long enough to do the work themselves. Better tools may help with that. They don't take the job off our hands.

Seeing the person. Education has always been about more than knowing things. It's also about attitudes, values and how we treat each other. Krathwohl, Bloom and Masia called this the affective domain in 1964. I've written curriculum with both kinds of aims, including for the Canadian Marketing Association's professional designation. Knowing a professional standard and practicing it are different things to teach.

Teaching changed my own attitudes, too. In business, I was used to attention following the hierarchy. As a teacher, I felt responsible for being fair, open and transparent with every student. I found that freeing.

That's where empathy comes into teaching, and it runs both ways. A student's experience can change what the teacher believes. A classmate's story can change the whole conversation. You don't only learn about empathy. You practice it with other people.

I don't want to draw too neat a line here, with machines for facts and people for feelings. The Harvard students said the AI tutor made them more engaged, and AI may help some people practice in private or find words for something hard. Some learners already treat it as a collaborator, one that pushes their logic and stretches what they understand, and they love it. But listening, caring and being changed by someone else still happen between people. If teachers become guides, that's where much of their time should go.

Most of the public argument is about kids in school. I think adults have just as much at stake.

The World Economic Forum's 2025 survey of employers expects 39% of workers' existing skills to change or become outdated by 2030. That's the employers' forecast, not a measurement, and it says nothing about teachers disappearing. It does give you a sense of how much learning working people will be expected to do.

Adults come to learning with jobs, families and not much time. They also bring years of experience, real problems and very little patience for anything that won't help them on Monday morning. For them, instruction that's there whenever they need it could make a huge difference.

But in a room full of working adults, the teacher is rarely the only one who knows something. Someone may understand how an idea has to work in their own job better than the teacher does. Their experience can change the question, or show why a perfectly good answer won't help. A teacher who asks for that experience has to be ready to hear something that changes their mind.

That's the guide I picture for adults, part coach and part mentor: someone who designs the experience, asks the question that matters and makes sure no one is learning alone. AI could make more of that possible. Someone still has to decide to make room for it.

I keep coming back to one class.

In the fall of 2017, after nine semesters teaching online, I taught in a classroom for the first time. It was the first course in the certificate. Friends who were instructors had pushed me to try it, and they were right. The room was in the OISE building on Bloor Street in Toronto, right above St. George subway station. Logistically, it was the worst classroom I've ever had.

The assignment was to tell a story from your own life using multimedia. One of my students was Shawn Ahmed, the son of Bangladeshi immigrants, a gay Muslim and an activist. After the Orlando nightclub shooting in 2016, he spoke out online and was threatened for it, which he wrote about in U of T Magazine. The following year, he told his story in our classroom.

When he finished, I was shocked and crying. I didn't want to cry, but I was. Others in the class were emotional, too. I told them I was gay as well. Our lives were very different, but I knew what it was to grow up feeling alone. Shawn had brought his life into the room, and I answered with a piece of mine.

That night is the best answer I have to the question this piece started with. A guide asks good questions. A guide also pays attention to the person, not just the work, and lets them see when something they've shared has landed.

I still think Aman is right. AI will do more and more of the teaching, and a lot of it will be done well. But ten years from the last lecture, I hope there are still rooms like that one, with a guide in each of them who is listening closely enough to be moved.

The AI Collective is built by volunteers across 180+ chapters in 40 countries.

Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.