Unpacking Vibe Coding: Help-Seeking Processes in Student-AI Interactions While Programming

A new paper from researchers at multiple European universities, accepted at the 27th International Conference on Artificial Intelligence in Education (AIED’26), unpacks the phenomenon of 'vibe coding': students writing programs by conversing with generative AI in natural language rather than writing code line-by-line. The study analyzed 19,418 interaction turns from 110 undergraduate students, applying inductive coding and Heterogeneous Transition Network Analysis to compare how top-performing and low-performing students used AI.

The results reveal a stark divide: top performers engaged in instrumental help-seeking, asking questions, exploring alternatives, and treating the AI as a tutor. AI responded with explanations and guided reasoning. Low performers, by contrast, leaned on executive help-seeking — they simply delegated tasks, prompting the AI to act as an executor and produce ready-made solutions. The key finding is that current generative AI mirrors the student’s intent, whether productive or passive, rather than optimizing for learning. The authors call for pedagogically aligned design that detects unproductive delegation and adaptively nudges students toward inquiry and cognitive effort.