26 March 2026

Why Your Students Still Need to Learn to Program (Even with AI)

Back

Written by

Becci Peters

Artificial intelligence is rapidly becoming part of students’ everyday learning. From generating code to writing essays, it can feel like a powerful shortcut.

But a recent research paper, How AI Impacts Skill Formation, raises an important question:

If AI helps students complete work, are they actually learning?

Let’s break down what the research says, and what it means for your classroom.

What the research did

Researchers ran a controlled experiment with software developers learning a new programming library.

One group used AI tools
One group did not
Everyone completed the same learning tasks and was then assessed

The key aim was simple:

Does AI help people learn new skills, or just finish tasks faster?

Key findings (in plain English)

Despite expectations, participants using AI were not much faster overall. Some even spent extra time prompting and interacting with the AI.

Those using AI scored significantly lower (around 17%) on assessments of:

Conceptual understanding
Code reading
Debugging

In short: they could produce answers, but didn’t understand them as well.

Participants who let AI do most of the work:

Finished tasks efficiently
But learned the least

The researchers identified different ways people used AI.

Learning was stronger when students:

Asked for explanations
Checked their own thinking
Engaged critically with outputs

Learning was weaker when students:

Copied answers
Used AI to “fix” work without understanding

Debugging: the biggest hidden risk

One of the most important and easily overlooked findings in the paper relates to debugging.

Debugging showed the largest drop in performance when AI was used.

Why did debugging suffer?

The research highlights three key issues:

1. Fewer opportunities to encounter errors

When AI generates working solutions:

Students don’t hit as many problems
They miss chances to practise troubleshooting

And without encountering errors, students don’t learn how to fix them.

2. AI becomes a “debugging crutch”

Some participants relied on prompts like:

“Fix this for me”
“What’s wrong with this?”

But instead of supporting thinking, this often replaced it.

Students delegated the reasoning, not just the task — and their learning suffered.

3. Reduced ability to diagnose problems

Debugging is more than fixing code — it involves:

Spotting that something is wrong
Understanding why
Testing possible solutions

Students using AI were less able to:

Identify errors
Explain what had gone wrong
Fix issues independently

Why this matters

This isn’t just about programming. Debugging is the skill students need to evaluate AI outputs. If students don’t develop it, they may:

Trust incorrect answers
Struggle to critique AI-generated work
Become overly dependent on AI

The big takeaway

AI can boost performance without building understanding, or even more bluntly: AI is not a shortcut to competence.

What this means for teachers

This research doesn’t say “don’t use AI.”

Instead, it shows:

How you design tasks matters more than whether AI is used.

Designing better AI-supported tasks

Here are some practical strategies you can use immediately:

Weak task:

“Write a program that…”
Stronger task:

“Explain how this program works and why it solves the problem.”

If AI can generate the answer, assess the understanding instead.

Ask students to:

Annotate AI-generated work
Identify errors or limitations
Rewrite outputs in their own words

This forces cognitive engagement, which the study found protects learning.

Encourage prompts like:

“Explain this step by step”
“Why does this work?”
“What would happen if…?”

Avoid:

“Do this for me”
“Fix this”

Given how important debugging is, deliberately design for it:

Include intentional errors in tasks
Ask students to diagnose problems before using AI
Occasionally ban AI for debugging stages

If students never struggle with errors, they never learn to fix them.

Include moments where students must:

Work independently
Complete retrieval practice
Demonstrate unaided understanding

This helps you see what they’ve actually learned.

Consider assessing:

Drafts
Prompt history
Reflections on AI use

Example: “How did AI help you, and what did you still need to figure out yourself?”

The research reinforces something teachers already know:

Struggle is part of learning.

If AI removes all challenge:

Students may complete tasks
But won’t develop durable knowledge

Questions to reflect on

When planning tasks involving AI, ask yourself the following:

What thinking do I want students to do that AI cannot replace?

Where might students be tempted to outsource their thinking?

How will I know if they truly understand?

Am I rewarding completion… or learning?

Where are students required to find and fix mistakes themselves?

Final thought

AI is a powerful tool — but like calculators or spellcheck, its impact depends on how we use it. This research is a timely reminder:

If students use AI to think less, they will learn less.

If they use AI to think more deeply, it can enhance learning.

The difference isn’t the tool, it’s the task design.

Read the full report here