Type: url
Original: https://arxiv.org/pdf/2608.20446
Status: generated
Summary
This state-of-the-art review assembles the first seventeen months of evidence on vibe coding from a cross-disciplinary corpus of 123 cited sources, covering the model and tool landscape, benchmark performance by task type, productivity, and risk. It argues that productivity figures that look contradictory (+26% tasks per week, a 19% slowdown, +441% review time) are consistent once measurement method, scope, and time horizon are held constant, and it identifies six patterns behind the dispersion. It further documents security failures, code-quality degradation, unsettled copyright exposure, and skill atrophy, and closes with the falsifiable conjecture that the gains are real on new code but shrink or reverse on mature codebases.