A Mixed-Method Empirical Study of LLM Assistance in Software Engineering Workflows

AI assistance changes how student developers approach a task, which complicates judging its value by generation speed alone. A software-engineering study surveyed 157 first- and fourth-year undergraduates, then studied 20 selected participants under assisted and unassisted conditions. Screen recordings showed assisted workflows often beginning with AI, transferring generated code, and returning to AI for debugging. Unassisted workflows relied more on documentation, existing templates, and trial-and-error refinement. The authors conclude that benefits depend on the task, experience, and verification practices. The evidence describes a small student experiment, so it does not establish an industry-wide productivity gain.