CodeExemplar: Example-Based Scaffolding for Introductory Programming in the GenAI Era

Researchers Boxuan Ma and Shinichi Konomi have introduced CodeExemplar, a novel AI-powered system designed to revolutionize introductory programming education in the age of generative AI. The core innovation addresses a fundamental tension: while tools like GitHub Copilot and ChatGPT can instantly generate working code, this creates a pedagogical dilemma where students might copy solutions rather than develop reasoning skills. CodeExemplar tackles this through example-based scaffolding, where the AI generates examples that match the underlying reasoning pattern of a target task but differ in surface context—supporting analogical transfer while reducing direct copying.

The system includes a two-dimensional taxonomy and concrete design guidelines for creating effective scaffolded examples. The researchers have built a functional prototype that integrates directly with auto-graded programming tasks, providing immediate, context-appropriate support. Initial validation comes from formative feedback gathered through classroom pilots and instructor interviews, suggesting the approach helps students learn programming concepts without relying on AI-generated solutions. This represents a significant shift from AI as a solution generator to AI as a pedagogical partner that guides rather than gives answers.

By focusing on the reasoning patterns behind programming problems rather than specific solutions, CodeExemplar helps students develop transferable skills that apply across different programming contexts. The system's integration with existing educational infrastructure means it could be deployed alongside platforms like Codecademy, Coursera, or university learning management systems. This research, published as arXiv:2603.23830, provides both theoretical framework and practical implementation for how AI can enhance rather than undermine programming education.