Egoist Learning
Back to Blog

Ai In Education Adaptive Learning Teaching Research Notes

What Teaching Becomes When Every Learner Has an AI Tutor

AI can give every learner personal support. I think that will make good teachers and mentors more relevant, not less.

By Paul Butad, Founder

I keep seeing new roles appear on LinkedIn because of AI. One of them stood out to me: Head of Education.

It makes sense. AI companies are building tools that explain, guide, answer questions, create learning paths, and give feedback. Whether they say it directly or not, they are entering education. But putting AI inside a learning product does not make it a tutor. Most of the time, it is still a chatbot with an education prompt. I know because I keep trying to use AI to learn things fast.

I had three days to learn how salespeople think

During a hackathon, I needed to learn sales in around three days. Not just a definition of sales. I needed to understand how salespeople work, how they make decisions, which tools they use, and what their day-to-day work actually looks like. I asked AI to help me. The first outputs were generic garbage.

It gave me information about sales. It did not help me learn enough about sales for the situation I was in. The hackathon was only a few days away. I did not have time for a complete beginner course or another list of books, frameworks, and sales terms.

The answers only became useful after I did the hard work of giving the AI my proper context:

  • Why I needed to learn sales.
  • What I needed to understand before the hackathon.
  • What I already knew.
  • What I did not know.
  • How much time I actually had.

That bothers me because giving perfect context is already difficult when you are learning something new. You do not know which part matters yet. You may not even know the right words for your gap. If the learner needs to act like an expert prompt engineer before the AI can teach them, then the AI is not an adaptive tutor yet.

An adaptive tutor needs two inputs

The first input is a basic understanding of how learning works. Retrieval is stronger than another round of rereading. Prior knowledge changes how much guidance someone needs. Feedback should help the learner see the goal, their current attempt, and the next useful correction. The learner should explain and apply the idea, not only read an AI-generated answer and feel smart for five minutes. These are not secret teaching tricks. We already have decades of learning research. The AI needs that foundation before it starts deciding how to help.

The second input is the learner’s context.

  • What are they trying to become capable of doing?
  • What can they already do?
  • Where are they struggling?
  • What have they tried?
  • What do they think the gap is?
  • What are their real limits right now?

I saw how much this changes advice when I recently asked a software engineering leader I look up to for career guidance. I honestly shared my struggles, my current level, and what I thought my gaps were. His advice was spot on because it was based on me. He even corrected the gaps I thought I had.

That last part matters. Personalization is not just remembering what the learner said. The learner can diagnose themselves incorrectly. A reliable tutor or mentor needs enough evidence and experience to say, “I think the real gap is somewhere else.”

A friendly teaching style is not enough

Once the AI understands learning science and the learner, it still needs to use that information properly. I do not want it to randomly choose a friendly teaching personality and call that personalization. “Explain like a patient tutor” is not enough.

The proof is whether the support changes after the learner does something. A learner who is confused should not get the same next move as someone who is ready for a harder challenge. If the AI only remembers your name and goal, that is personalized chat. It is not adaptive tutoring.

Some teaching rules also need to be mathematically encoded. They cannot all depend on whatever response the AI feels like generating next. The system needs a consistent way to interpret learner signals, adjust its support, and learn from what happened. Otherwise, the same learner can give the same evidence and receive completely different guidance because the model changed its wording or reasoning.

This does not mean reducing a learner to one score. It means giving the AI enough structure to be consistent while the learner keeps changing.

Teachers become more relevant

This is where I disagree with the fear that AI tutors will make teachers, mentors, or coaches irrelevant. I think they become more relevant than before.

AI can collect the learner’s answers, organize their history, adjust practice, and prepare a possible next move. A good teacher or mentor can look at the same learner and see something the system missed. They can correct the learner’s self-diagnosis. They can challenge a goal that is too safe. They can notice when the learner is avoiding the real problem. They can use experience from situations the AI has never lived.

I saw a small version of this while helping my sister move toward analytics engineering. My first instinct was to tell her to study one of the tools used in that work. It sounded like the correct place to start. But when we used our own product, Egoist Learning, the starting point changed.

She first needed to understand the problem in the manual data pipeline we already had. Why was the manual work painful? Why did it need to be automated? The tool only made sense after the problem became clear.

Lol, yes, we use our own product for our own benefit. The AI did not replace me in that moment. It helped us slow down and see a better starting point. I still brought my experience. She still needed to do the learning. The tool made the support more personal.

That relationship is what I want: AI prepares and adapts. Teachers and mentors judge, correct, challenge, and care.

Maybe you were never bad at learning

This is the outcome I care about, especially for younger learners. A learner struggles inside a one-size-fits-all system and starts believing they are bad at learning. The lesson moves before they understand the prerequisite. The same explanation is given again, only louder or slower. Their grade becomes proof that something is wrong with them.

Maybe the learner was not the problem. Maybe the goal was unclear. Maybe the starting point was wrong. Maybe the explanation did not fit. Maybe the learner needed a smaller challenge, a different example, faster feedback, or one person who understood what they were trying to do.

I want more people to become addicted to learning the way they get addicted to games or entertainment. Games keep showing you where you are, what changed, what challenge comes next, and whether your last move worked. Learning often gives you one score after everything is already over.

An adaptive tutor could help change that. It could make progress clearer, feedback faster, and the next challenge better matched to the learner. It can also go badly. AI can give wrong advice with complete confidence. Learners can depend on it instead of thinking. Private learning data can be collected and abused. Good tools may only reach people who can afford them. These risks are real. So is the opportunity.

We should ask harder questions than, “Will AI replace teachers?”

Who teaches the AI how learning works? Who checks the learning framework? Who corrects the AI when it misunderstands the learner? Who decides what information the system should never collect? Who helps the learner choose a goal worth reaching? Those sound like jobs for educators.

Research notes

The ideas in this note connect research we have already used while building Egoist Learning: