Vibe Coding: Top Students Ask Questions, Low Performers Delegate to AI
Researchers analyzed 19,418 AI-help interactions from 110 undergrads learning to code. They found a sharp split: high performers used AI as a tutor (asking questions, exploring), while struggling students treated it as an executor (dumping tasks, expecting finished code).
• Top performers: instrumental help-seeking (inquiry, debugging logic together)
• Low performers: executive help-seeking (delegation, copy-paste workflows)
• Pattern held across 110 students — statistically robust distinction