The current impact of AI on engineering velocity (newsletter.getdx.com)

During a recent session at DX Annual, Abi Noda, CEO of DX, and Brian Houck from Microsoft unveiled new research on the impact of AI on engineering velocity. The findings indicate that while AI adoption has increased pull request (PR) throughput by 10 to 15 percent for most organizations, these gains are significantly less than the tenfold improvements often touted in industry narratives. This research reveals that coding, which constitutes only about 14 percent of a developer's workload, cannot explain the overall productivity boost; significant bottlenecks remain in planning, reviews, and testing. The study highlights the nuanced relationship between AI-driven efficiency and actual engineering productivity. Although AI can streamline some aspects of coding, it often brings new bottlenecks in quality assurance and oversight. Additionally, the concept of "false velocity" warns that an uptick in code output doesn't necessarily translate to greater business value, as it may lead to increased technical debt and maintainability challenges. As developers face cognitive debt—where they produce code faster but understand the systems less well—organizations are now focusing on applying AI across the entire software development life cycle (SDLC). With evolving measurement frameworks, leaders are starting to consider the distinction between human and autonomous productivity to better evaluate the full impact of AI on engineering practices.