Eight Myths on Software Engineering and GenAI
Points and comments are a snapshot, not live.
Eight persistent myths about GenAI in software engineering, grounded in research evidence.
An ACM Queue article debunks eight myths about generative AI in software engineering. Drawing on large-scale studies, it argues developers spend only ~14% of their time coding, making code-generation a narrow productivity lever. Lines of code remains an invalid metric, and AI does not help all tasks or engineers equally: gains depend on task type, developer experience, and context. Productivity requires organizational redesign, not just tool access. Adoption faces barriers of trust, competence penalties, and workflow integration. Enterprise AI speed cannot match startups due to legacy systems, compliance, and reliability constraints.
What commenters are saying
Commenters broadly agree with the article's premises. Top-ranked comments highlight that coordination, not coding, is the true bottleneck in most development pipelines. Several developers report that AI has shifted their time: they now spend more of it reviewing and orchestrating agent-generated code. A debate emerges over whether to review every line of AI code, with some advocating selective trust for straightforward output and others warning of security risks (e.g., SQL injection from string concatenation). One commenter notes the challenge of reviewing 16k lines across three apps due to AI code bloat.