Misryoum reports on new evidence that the best AI tutors may be the ones that adapt practice difficulty—keeping students in the learning “sweet spot.”
For a moment, the promise of an AI tutor sounds simple: ask a question, get an answer, move on.
But Misryoum has learned that education researchers are finding a more complicated truth—chatting isn’t the hardest part of tutoring. The harder part is deciding what the student should do next.
The shift from “explanations” to practice sequencing
The latest push comes from a study involving nearly 800 Taiwanese high school students learning Python programming with an AI tutor. Everyone used the same chatbot-based system and, crucially, it was designed not to hand over full solutions.
Where the learning experience diverged was in the path the students took through practice.. Half of the students moved through a fixed sequence of problems, gradually increasing in difficulty.. The other half received a personalized sequence. with the system adjusting what came next based on how students were performing and interacting inside the course.
That approach is rooted in a familiar teaching concept—keeping learners in the “zone of proximal development.” In plain terms. tasks that are too easy don’t pull students forward. while tasks that are too hard can turn learning into frustration.. The goal is a narrow middle ground: challenging enough to drive effort. but not so steep that students give up or disengage.
What personalization looked like in the study
Misryoum notes that personalization here wasn’t just about tone or wording. The AI wasn’t only responding to students’ questions; it was also shaping the learning journey itself.
A team at the University of Pennsylvania tested the system using signals such as how students answered practice questions, how often they revised or edited their code, and how they interacted during conversations with the tutor. Those patterns were then used to select the next problem.
The study reported that students in the personalized group performed better on a final exam than those assigned to the fixed problem order.. The research team described the size of the difference as being comparable to many months of additional schooling. though the work was presented as early evidence and Misryoum understands it had not yet gone through peer review.
Why this matters for families—and for classroom reality
The most immediate impact of AI tutoring isn’t academic theory—it’s time and motivation. In real school life, students don’t just need explanations; they need practice that keeps them working.
The study suggested engagement may have been the mechanism.. Students in the personalized condition spent more time practicing overall—turning what could have been a short after-school module into a longer. more sustained effort.. That matters because coding, like many skills, doesn’t absorb itself through reading alone.. Students learn by trying, failing, revising, and trying again.
Misryoum also sees a practical equity angle here.. The personalized approach appeared to help students who were newer to Python more than those who already had coding experience.. The findings also indicated benefits for students from less elite high schools.. That doesn’t mean the technology solves all gaps. but it suggests adaptive sequencing could be one lever for reducing the “wrong level of challenge” problem that quietly undermines learning.
The caution Misryoum can’t ignore: AI tutors can backfire
Even with promising results, Misryoum is cautious about the broader AI tutor trend. Earlier research and experiments have raised concerns that students may lean too heavily on chatbots, accept spoonfed solutions, and fail to internalize the underlying material.
The difference in this study is that the system was intentionally built not to provide direct answers. In other words, personalization was paired with a design that pushes students back toward problem-solving rather than dependency.
Researchers also acknowledge that students often don’t know what they don’t know. A tutor that answers questions politely can still miss the deeper need—because asking better questions requires a foundation the learner may not have yet.
That’s why the study’s emphasis on “what to practice next” is so significant: it moves support from the student’s ability to request help to the system’s ability to detect learning needs.
Intelligent tutoring systems: the lesson from the past
Misryoum also recognizes that this idea isn’t brand-new.. Before modern generative chatbots. education researchers built “intelligent tutoring systems” that attempted to estimate what students knew and present the next best problem.. These systems often relied less on natural conversation and more on hints, feedback, and calibrated practice.
Where earlier tools struggled wasn’t learning science—it was adoption and engagement. Many students didn’t want to use them.
Today’s AI tutors can be more engaging precisely because they can converse in a way that feels responsive.. But the key question is what students do during that engagement.. If conversation simply becomes entertainment or a shortcut, learning suffers.. If conversation becomes the wrapper around well-timed practice, the technology can start to earn its place.
Humans still have a role—especially for struggling students
One future direction being explored by researchers is a hybrid model: AI that detects when a student is drifting, combined with human support to bring students back.
Misryoum sees the logic in that.. Even a well-designed adaptive system may not know how to motivate every learner in every moment—especially when students are disengaging silently rather than struggling openly.. A human tutor can respond to mood. persistence. and effort in ways that a software model often can’t fully capture.
As AI tools become more capable, the emerging consensus is not that teachers will be replaced. It’s that tutoring will become more targeted—less one-size-fits-all, more responsive to individual progress.
For now. the most actionable takeaway from this study is surprisingly old-fashioned: learning improves when students are kept in the right zone of difficulty and given enough practice to turn effort into skill.. The difference is that AI may now help keep that balance automatically—problem by problem.
Peer & Self Assessment Boosts Student Motivation, New Study Suggests
11+ Tuition Reframed: Curriculum-Led Learning, Not Rehearsal
A new need-to-know for the AI classroom: 5 launch activities