Type: url
Original: https://arxiv.org/pdf/2605.23135
Status: generated
Summary
This longitudinal mixed-methods study surveyed professional software engineers using AI coding assistants twice, six months apart (158 and 101 eligible participants, 95 matched), about task focus, developer experience, and productivity. It reports a shift from creation toward verification work, proposes a new category it calls supervisory engineering work (directing, evaluating, and correcting AI output), and identifies a productivity-experience paradox in which perceived productivity stayed stable while reported developer experience, especially flow state and cognitive load, eroded.