Understanding Github Copilot Adoption In Software Engineering Education: An Exploratory Repository Mining Study

The integration of AI-powered programming assistants into software development workflows has begun to influence how students design and implement code. This paper presents an exploratory study on the use of GitHub Copilot in collaborative academic projects developed by third-year undergraduate Computer Science students at the University of Costa Rica during the 2025 academic year. Leveraging a custom Commit Mining System, we analyzed fine-grained metadata which explicitly attributes contributions to AI tools in commit histories. Our methodology utilizes ChatGPT 5.2 to perform an automated intent detection on both commit messages and code diffs, classifying 433 AI-mediated commits across ten functional categories. Results indicate a concentration of AI involvement in maintenance and verification tasks rather than core logic; documentation was the most frequent category, accounting for 32.1% and 38.7% of AI activity in the studied repositories. Conversely, core logic implementation remained low, with Feature commits representing less than 5% of the classified contributions. The findings indicate a shift from traditional "read → understand → implement" cycles to a GenAI-mediated workflow characterized by a "prompt → response → implement" pattern. High frequencies of Fix (20%) and Style (21.8%) commits suggest that students frequently encounter the "productivity-quality paradox (a trade-off between productivity and quality)", where initial speed gains are offset by a necessary iterative refinement and debugging phase. These empirical insights highlight the challenges of ”cognitive disengagement” and the need for new pedagogical strategies that prioritize metacognitive steering, problem decomposition, and rigorous verification over traditional syntax mastery.