AI's 70% Problem

Addy Osmani, from Google's Chrome Developer Experience team, discusses the "70% problem" in AI coding during the Agentic Engineering Sessions. He highlights that while AI tools can generate up to 70% of code efficiently, the remaining 30%—involving edge cases, security, and production integration—remains challenging, often negating the initial productivity gains. Despite increased adoption, trust in AI-generated code is declining, with many developers reporting low confidence in its reliability. Osmani emphasizes the importance of human oversight, urging engineers to understand and review AI-generated code to ensure security and maintainability. He also notes that productivity gains touted by some companies are often context-dependent, particularly in greenfield projects lacking technical debt. Osmani advises junior developers to use AI as a learning tool while maintaining curiosity and a commitment to understanding the code they work with.