January 30, 2026
Generative Artifical Intelligence (AI) is rapidly transforming programming education. Particularly, AI assistance has been shown to accelerate task completion and enhance motivation – whether these gains also translate to learning outcomes has been under debate.
A team of researchers from the Technical University of Munich, led by Patrick Bassner with EdTech Executive Board Member Stephan Krusche and EdTech Fellow Ben Lenk-Ostendorf among its members, conducted a large-scale study in a university setting, providing new empirical evidence that AI can improve task performance while showing no significant differences in conceptual understanding compared to traditional resources.
Empirical Study
275 students from the Technical University of Munich worked on a programming task under one of three conditions:
- AI tutor (Iris), that provided hints but withholded full solutions;
- Unrestricted AI (ChatGPT), that could provide complete answers on demand;
- No-AI control group, that only allowed access to traditional web resources.
Results
The study investigated how GenAI support affects programming education, in particular learning outcomes, cognitive load, frustration, and motivation.
Learning Outcomes:
Both AI-supported groups achieved significantly higher scores in the programming exercises than the control group. Yet these performance gains did not translate into greater conceptual learning or code comprehension abilities.
Cognitive Load & Frustration:
Students using AI reported lower frustration and reduced cognitive load.
Motivation:
Solely the AI tutor increased intrinsic motivation, whereas unrestricted AI use encouraged a “comfort trap”. Despite students associating ease and helpfulness with unrestricted use of AI, the feeling of smoother progress by having immediate access to solutions only hides weaker engagement with core concepts and less learning. It is the constructed, hint-based system that preserves a unique balance between guided support and cognitive challenge.
Takeaway Message
As GenAI becomes increasingly embedded in educational contexts, the challenge is no longer whether to use AI, but how to design AI-supported learning in pedagogically meaningful ways. While unrestricted AI may undermine the productive struggle necessary for deeper learning, didactically grounded AI integration such as carefully constructed AI tutors can support students without replacing their cognitive efforts.
References
- Bassner, P., Lenk-Ostendorf, B., Beinstingel, R., Wasner, T., Krusche, S. (2026). Less stress, better scores, same learning: The dissociation of performance and learning in AI-supported programming education. Computers and Education: Artificial Intelligence, Volume 10, https://doi.org/10.1016/j.caeai.2025.100537
