During the first semester of 2026, I had the opportunity to teach an introductory programming course at university. Like many instructors, I knew artificial intelligence was going to be part of the classroom experience. What I didn't expect was how clearly it would expose the difference between completing work and actually learning.
One observation stood out above everything else.
Two completely different students
Throughout the semester, students submitted programming assignments to be completed at home. Many of them consistently scored between 90 and 100.
Then came the in-person assessments.
These exams were completed on paper, without computers, without IDEs, and without AI assistants. The exercises weren't designed to trick anyone. In many cases they were analogous to previous assignments, and occasionally they were almost identical with small variations.
The results were shocking.
Some students who consistently earned perfect scores at home suddenly scored between 30 and 40 in person.
This wasn't an isolated incident. It happened repeatedly.
As an instructor, it forced me to ask a difficult question:
Were students learning to program, or were they learning how to obtain solutions?
AI isn't the problem. Avoiding struggle is.
It's tempting to blame artificial intelligence.
I don't think that's the right conclusion.
The real issue is that AI makes it incredibly easy to skip the most important part of learning: struggling with a problem.
Learning programming has never been about reading the correct answer. It's about spending time being confused, making incorrect assumptions, debugging your own reasoning, and finally reaching the solution after multiple failed attempts.
That process is uncomfortable.
It's also exactly where learning happens.
When students immediately ask an AI to solve an exercise, they bypass the cognitive effort required to build mental models. They receive a correct solution, but their brain never had to construct one.
As a consequence, when faced with a similar problem days or weeks later—without AI—they often cannot reproduce the reasoning that led to the answer.
The solution was never really theirs.
Not every AI user performed poorly
Interestingly, the highest-performing students also used AI.
The difference wasn't whether they used it.
It was how they used it.
The students who consistently performed well in both take-home assignments and in-person exams tended to use AI as:
- a tutor when they got stuck,
- a tool for generating study summaries,
- a way to create additional practice exercises,
- an extension of the classroom when a concept wasn't completely clear.
Crucially, they still attempted the problems themselves first.
They argued with the AI.
They asked why.
They compared different approaches.
In other words, they remained the primary problem solver.
These students generally showed noticeably stronger understanding and retained concepts far better during in-person evaluations.
The assessment problem
This experience convinced me that AI has not fundamentally changed learning.
It has exposed weaknesses in how we measure it.
For decades, many assignments assumed that work completed outside the classroom reflected individual understanding. That assumption is becoming increasingly fragile.
If an assessment primarily measures whether a student can produce an answer with unlimited external assistance, it may tell us very little about what the student actually knows.
This doesn't mean we should ban AI.
It means we need better assessments.
Programming courses—and honestly, many disciplines—must evolve beyond grading the final artifact alone.
We should value:
- reasoning over answers,
- iteration over perfection,
- explanation over reproduction,
- problem-solving over solution copying.
The objective is no longer simply asking, "Can the student produce code?"
The more important question is:
"Can the student explain why this solution works, adapt it to a new situation, and build it again without assistance?"
Teaching students how to use AI
Perhaps the biggest lesson I learned is that we should spend less time telling students not to use AI and more time teaching them how to use it.
AI can be an incredible teacher.
It can explain concepts from different perspectives.
It can generate infinite practice exercises.
It can summarize lectures.
It can provide immediate feedback.
It can answer questions at 2 a.m. before an exam.
But none of those benefits matter if the student delegates the thinking itself.
The objective should never be replacing the learning process.
It should be supporting it.
Final thoughts
Artificial intelligence isn't making students incapable of learning.
It is making it much easier to avoid learning.
Those are very different things.
As educators, our challenge is no longer deciding whether AI belongs in the classroom. That debate is already over.
Our responsibility is to redesign learning experiences so that students still have to wrestle with ideas, confront uncertainty, make mistakes, and develop the reasoning that no language model can permanently transfer to them.
Because at the end of the day, education has never been about producing correct answers.
It's about producing people capable of finding them.
Top comments (0)