Hutchison, Applebaum, Angelikas, Patel et al. (2026) — cs.HC (ITiCSE 2026) 📄 Full text (arXiv)
Hutchison et al. (2026) develop and validate a method for measuring critical engagement with AI code completion tools in educational settings. Using behavioral signals (time-to-accept, edit distance from suggestion) and embedded attention checks, they find that the majority of students accept AI code suggestions passively, without critically evaluating correctness or appropriateness. This 'tab-and-go' behavior directly threatens the development of programming skills, as students bypass the cognitive effort required for ai-learning-transfer. The work provides a methodological toolkit for formative-assessment in AI-augmented programming courses, enabling instructors to detect when students are over-reliant on AI. The findings connect to student-experience research in stem-education by showing that the mere availability of AI tools does not lead to productive learning — structured pedagogical interventions are required to ensure students engage critically rather than deferring to AI output.