Fostering Metacognitive Skills in Programming: Leveraging {AI} to Reflect on Code

The increasing use of AI among students has significant implications for established practices across all disciplines. In the specific case of programming in Computer Science (CS) education, we are observing a debate between the education system which sees AI-generated code as a threat to the learning’s quality, and the industry, which expects professionals to best take advantage of AI-assisted programming. In this context, a successful mediation lies in fostering skills such as metacognition and reflective learning to bridge the academic and professional worlds. This paper reviews the literature on AI-assisted practices supporting metacognition and reflective learning. Drawing on this review and the findings from a prior pilot study run by us, we designed the Reflective AI Programming Lab (RAP Lab) where groups of three students collaborate to solve programming tasks using exclusively AI-generated code, with restricted queries and a set of constraints on a designated platform which logs all the interactions between students and the AI. The approach leverages AI-driven feedback and collaboration enhancing dialogical practices as Pair Programming and promoting the development of critical reflection on AI tools in CS. By having students explain in detail their reasoning and structure their solution strategies to a third party (AI), this intervention stimulates metacognition and reflective learning by offering a different perspective on problem solving. In fact, this approach promotes a deeper comprehension of the problem and forces students to clarify and refine their thoughts when articulating their solution strategy. AI serves as an impartial non-judgmental observer, allowing students to explore their mistakes without fear of embarrassment, encouraging a risk-free environment where they are more likely to experiment, learn from their errors, and engage in deeper reflective learning. Although this approach has yet to be validated, it will serve as the basis for more extensive data collection with a larger sample in the upcoming semester.