DX research drawing on real customer data shows AI coding tools are delivering modest engineering productivity gains — typically 8–15% increases in PR throughput — far below the 10x claims common in industry headlines. Key reasons include coding representing only ~14% of developer time, new bottlenecks created by faster code generation (reviews, QA), social friction, and AI tools lacking codebase context. The discussion warns against 'false velocity' where more PRs don't translate to business value, introduces 'cognitive debt' as a risk where developers ship code without fully understanding it, and argues the biggest AI opportunity lies in the other 86% of engineering work: planning, documentation, incident response, and autonomous agents working in parallel with humans. Measurement frameworks are also evolving to separately track human acceleration vs. agent-performed work.
