A randomized controlled trial with 52 software developers found that using AI coding assistance led to 17% lower quiz scores on concepts just learned, equivalent to nearly two letter grades, though task completion was only slightly faster. The study identified distinct interaction patterns: low-scoring developers delegated heavily to AI, while high-scoring ones used AI for comprehension through follow-up questions and explanations. Results suggest aggressive AI adoption may hinder skill development, particularly debugging abilities, even as it boosts productivity on familiar tasks. The findings highlight the importance of intentional AI tool design and usage policies that balance efficiency with learning.
415 Impressions