A Systematic Review and Taxonomy of GenAI Usage in Programming Education: A Self-Regulation Perspective
This systematic review of 58 studies up to October 2025 proposes a four-dimensional taxonomy to demonstrate that the impact of Generative AI on student self-regulation in programming education depends not on the technology itself, but on structured pedagogical mediation strategies that foster autonomy versus unguided contexts that encourage cognitive dependence.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are learning to ride a bicycle. In the past, you had to wobble, fall, and figure out how to balance all by yourself. Now, imagine a magical, invisible training wheel that can talk to you. It can tell you exactly how to pedal, steer you away from potholes, and even fix your bike if you crash. This is what Generative Artificial Intelligence (GenAI) feels like in the world of learning to code. It's a super-smart computer program that can write computer instructions (code) for you, explain complex ideas, and help you solve problems instantly.
But here is the tricky part: Self-Regulated Learning. This is a fancy term for "learning how to learn." It's the ability to set your own goals, keep yourself focused when things get hard, and check your own work to see if you really understand it. The big question isn't just "Can this magic tool help me finish my homework?" but "Will using this tool make me a better, more independent learner, or will it make me dependent on the machine?" If the tool does all the thinking for you, you might get the answer right, but you won't learn how to ride the bike on your own. This is the exact puzzle scientists are trying to solve in the field of Programming Education.
The Great Code-Helper Investigation
A team of researchers decided to take a giant step back and look at the whole picture. They didn't just run one experiment; they acted like detectives, hunting down 58 different studies published up to October 2025. They scoured through thousands of research papers to find out: How are teachers and students actually using these AI tools to learn coding, and does it help them become independent thinkers or just code-copying robots?
They found that the answer isn't a simple "yes" or "no." It's more like a recipe. The magic tool (GenAI) is just an ingredient; what matters is how you cook with it.
The Four Ingredients of the Recipe
To make sense of all the different ways people are using AI, the researchers created a special map called a Taxonomy. Think of it like a menu at a restaurant that helps you order the right meal. They organized the studies into four main categories:
- What are we cooking? (Target Syllabus): Are we using the AI to write new code from scratch, fix broken code (debugging), or just explain how things work? Most studies focused on writing code, which makes sense since that's what tools like ChatGPT are best at.
- Who is the chef? (Mediation Strategy): This is the most important part. Is the AI just sitting there waiting for you to ask it anything (Free Use), or is the teacher guiding you on exactly when and how to use it (Guided Use)?
- The "Free Use" Trap: When students are left alone with the AI, they often treat it as a shortcut. They ask for the answer, get it, and move on. The researchers found this leads to a "illusion of competence"—you think you know how to ride the bike because the invisible wheel held you up, but you actually don't.
- The "Guided" Magic: When teachers set rules—like "Use the AI to explain the error, but don't let it write the fix for you"—students actually learn more. It's like having a coach who won't push the bike for you but will tell you which way to lean.
- How do we feel? (Cognitive Impact): Did the students feel more motivated? Did they understand the material better? Or did they become dependent on the tool? The results were mixed. Some students felt less stressed and more confident, while others felt they were losing their ability to think critically.
- What kind of tool is it? (Technical Approach): Is it a simple chatbot you talk to, a smart tutor built into your coding software, or a specialized robot? The study found that simple chatbots are the most common, but specialized tools designed by teachers often work better for deep learning.
The Big Discovery: It's All About the Rules
The most exciting finding of this paper is that the technology itself isn't the hero or the villain. The AI tool is neutral. It's the pedagogical mediation—the way teachers and students use it—that decides the outcome.
- When it goes wrong: If you just let students use AI without any rules, they tend to become "cognitive dependents." They stop trying to solve problems themselves because the tool is too easy to use. They might get good grades, but they aren't actually learning.
- When it goes right: When the AI is used as a scaffold (like training wheels that you slowly remove), it's amazing. Teachers who use strategies like "ask the AI for a hint, not the answer" or "use the AI to check your logic before you write code" see students become more motivated, more engaged, and better at solving problems on their own.
What's Missing?
The researchers also pointed out some holes in the current knowledge. Most of the studies they looked at were short-term (just a few weeks or even a single class session) and involved small groups of students. We don't really know yet if using AI for a whole semester makes students better coders in the long run, or if it just makes them faster at copying. Also, most studies focused on students who were already pretty good at coding; we don't know enough about how this affects beginners or students with different learning needs.
The Bottom Line
This paper tells us that Generative AI is a powerful new tool in the classroom, but it's not a magic wand that fixes everything on its own. If you just hand a student a magic wand and say "do your homework," they might get the answer, but they won't learn the magic. However, if a teacher uses that wand as part of a structured lesson—teaching students how to ask the right questions and when to put the wand down—then it can help them become independent, confident, and creative problem solvers.
The future of coding education isn't about banning AI or letting it run wild; it's about finding the perfect balance where the AI supports the student's brain without replacing it.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.