Eight Myths on Software Engineering and GenAI

Six Microsoft researchers and one University of Victoria professor systematically dismantle eight persistent myths about generative AI in software engineering, drawing on large-scale studies, interviews, and field observations. The myths span developer time allocation (only ~14% of time is spent coding), the fallacy of lines-of-code metrics, the context-dependence of AI productivity gains, the structural barriers to adoption, and why enterprises can't innovate at startup speed. The core argument: AI's benefits are real but overstated and misunderstood — context (task type, developer experience, team dynamics, organizational systems) determines outcomes far more than tool capability alone. The paper is a research-backed corrective to marketing claims, misread studies, and the "10x developer" narrative, arguing that realizing AI's value requires moving beyond hype and simplistic metrics toward the broader goals of secure, maintainable, high-quality software.