Students’ Perception Accuracy of Partners’ AI Use and its Relation to Collaboration Performance

Collaborative programming is a fundamental component of software engineering education, requiring students to coordinate tasks and align on design decisions. The recent, widespread adoption of generative AI tools has introduced a new, often invisible layer to this collaboration. Because AI usage is not always observable, students must infer their partner's reliance on these tools to effectively divide work and provide support. This study investigated how accurately students perceive their partners' AI usage and whether these perceptions influence team performance.

Researchers conducted a three-wave longitudinal study of 103 student pairs in an introductory software engineering course. Over the course of a one-month project, students reported their own AI usage and estimated their partner's usage across both general volume and specific programming tasks. The researchers analyzed whether these perceptions converged over time and how misalignment in these beliefs correlated with final project grades.

The study revealed that while many teams improved their perception alignment during the project, a significant portion remained misaligned until the end. Crucially, early misalignment in perceptions of AI usage was a significant predictor of lower project outcomes. The researchers found that students with lower prior programming performance suffered the most from these misaligned perceptions, suggesting that these students may be less equipped to navigate the coordination challenges posed by opaque AI usage. Furthermore, the study noted that face-to-face pair-programming sessions did not consistently resolve these perception gaps, indicating that simple interaction is insufficient to foster transparency.

As generative AI becomes standard in educational and professional programming, the "invisibility" of AI assistance poses a risk to team coordination. When partners cannot accurately gauge each other's expertise or reliance on AI, they may misallocate tasks or fail to provide necessary support. These results suggest that educators and tool designers should prioritize features that make AI-assisted workflows more transparent to collaborators, particularly for students who are still developing their foundational programming skills.