New prompt pattern framework prevents AI delegation, keeps students thinking in secure coding education

A new academic paper from researchers Haindl, Eigner, and Kieseberg tackles a growing problem in AI-augmented education: students using large language models to bypass the effortful engagement necessary for deep learning. The paper, published on arXiv, introduces a practical framework built on nine prompt engineering patterns synthesized from existing computer science literature. These patterns are mapped to two pedagogical constructs—Productive Struggle (students grappling with challenges) and Evaluative Judgement (students assessing their own work). The goal is to keep GenAI in the course without removing cognitive demands.

The framework is demonstrated through an Advanced Secure Coding module structured with the DELTA framework. Three specific patterns—Flipped Interaction (AI asks questions, students answer), Alternative Approaches (AI suggests multiple solutions, students compare), and Cognitive Verifier (students verify AI’s reasoning)—guide vulnerability discovery and remediation while keeping students in the reasoning role. The authors argue that this structured, replicable approach gives instructors fine-grained control over how students interact with AI, preserving student reasoning. It establishes a foundation for future empirical evaluation in live course settings, offering a practical way to design AI-augmented learning experiences that prioritize understanding over delegation.