Generative AI is reshaping software engineering, yet the narrative has outpaced evidence. Marketing claims, anecdotal wins, and misread studies have seeded persistent myths that quietly steer poor decisions about adoption, tooling, and measurement. This article dissects eight common misconceptions to provide a research‑backed foundation.

Myth 1: Developers spend most of their time coding. A 2025 study of >450 engineers at Microsoft found they write code only 14% of the day, with the rest devoted to design, meetings, and reviews. Even the best AI assist touches a surprisingly small slice of work.

Myth 2: Writing code is the bottleneck. If coding is only 15% of a developer’s effort, an AI that doubles speed could lift overall productivity by less than that share, while the remaining 85%—design, testing, integration—remains unchanged. Myth 3: Lines of code is the best metric. Bill Gates warned against this in 2014, and studies still see the metric rewarded, yet it misrepresents success and can encourage toxic behaviour.

The reality is messier: AI helps some tasks, developers, and contexts more than others. Productivity gains require workflow rethinking, trust, and learning time. The “startups move fast” narrative ignores enterprise constraints. Ground decisions in evidence, not hype.