The rise of Generative AI and its heavy usage by students have caused many educators to return to traditional supervised exams, even if this has potential pedagogical disadvantages. It is important to find forms of assessment which avoid the disadvantages of the exam, yet are robust against AI cheating, and possibly also facilitate student learning with AI where this is appropriate. The course IT1001 at the NTNU uses self-paced mastery learning combining frequent tests and incremental project work. This may be a course design enabling robustness against AI and constructive usage of AI, as the supervised tests work as a gatekeeper against AI cheating, while the project allows students to creatively explore AI usage. However, the AI usage as so far been rather ad hoc, with limited learning resources. This paper seeks to analyze whether students consider AI a useful learning tool in the course, and what tasks they have been using AI for. Results indicate that they consider AI as a useful tool, but less so in 2025 than in 2024, and not outcompeting the learning resources provided by the teacher. AI has been used for several different tasks related to programming, but not so much for massive outsourcing of work in terms of code generation. Rather, the most frequent types of usage were error finding and idea generation. The paper concludes by outlining how the coverage of AI can be improved for the next run of the course.
