Abstract:

Integrating generative artificial intelligence (AI) into higher education is transforming how students approach programming and algorithmic problem-solving. Previous work in our context has shown that integrating Generative Artificial Intelligence (GenAI) tools into advanced programming activities can improve performance, efficiency, and student experience when used under responsible classroom guidelines. In this work, the focus shifts from whether GenAI can help to how it should be taught and assessed so that it supports deep learning without encouraging uncritical acceptance of outputs or weakening academic integrity.

We designed and evaluated a combined pedagogical and assessment approach for postgraduate programming education aligned with authentic development workflows (e.g., notebook-based experimentation and integrated development environments). The instructional component adds explicit scaffolding for verification and correction: students follow structured routines to formulate requests, inspect assumptions, test generated code, diagnose failures, and document the reasoning behind fixes and design choices. GenAI suggestions are treated as hypotheses to be validated, not answers to be copied. In parallel, assessment was redesigned to be GenAI-aware and evidence-centered. Tasks require intermediate artifacts (e.g., test cases, debugging traces, short explanations of trade-offs, and change logs), and grading criteria place substantial weight on reasoning quality, correctness under constraints, and the ability to identify and repair issues introduced by GenAI-assisted code.

The evaluation uses a mixed-method design combining outcome measures (grades and rubric-based code quality), process indicators (time-on-task and frequency of external documentation lookups), and learner perceptions (self-efficacy, perceived cognitive load, and satisfaction). These data are complemented by traceability artifacts produced during assignments and brief technical defenses focused on explaining decisions and debugging strategies. Results indicate that the combination of verification scaffolding and GenAI-aware assessment strengthens students’ critical engagement with GenAI outputs, reduces overreliance on generated solutions, and maintains or improves learning outcomes while making tool use transparent and accountable. The study provides transferable design principles for programming courses adopting GenAI: teach verification explicitly, assess the reasoning process (not only the final code), and require traceable evidence of understanding.

Keywords:

Generative AI, Programming Education, Artificial Intelligence Learning, Active Learning, Ethical AI Integration.