The Impact of AI-Powered Coding Tools on Students' Learning Habits and Programming

AI coding assistants can generate code, explain programming concepts, and suggest debugging steps, but their educational value cannot be judged from task completion alone. This focused narrative review examines how AI-assisted programming relates to students’ learning behavior, programming performance, independent problem-solving, and psychological processes. It synthesizes recent empirical studies and reviews, including a meta-analysis of 35 controlled studies, a systematic review of 119 papers on AI in programming education, a quasiexperiment with 82 college students, a study of 17 novice programmers, and a randomized controlled trial with 275 students. Across this evidence, AI assistance is often associated with improved performance or efficiency on supported tasks, but gains in conceptual understanding and independent learning are less consistent. Behavioral Science helps explain this distinction through cognitive load, programming self-efficacy, frustration, intrinsic motivation, help-seeking, and metacognitive monitoring. A recent trial found that AI-supported groups performed better on a programming exercise than a no-AI group, yet did not show greater short-term knowledge gains or code-comprehension scores. The review argues that the instructional design of AI support matters: hint-first scaffolding, code verification, explanation, retrieval practice, and independent transfer tasks are more defensible than unrestricted answer generation as default teaching strategies. Because this paper is a focused narrative review, it does not claim exhaustive systematic coverage or report original student data.