The 55.8 Percent Productivity Number From Doshi And Vaishnav Is Narrower Than People Think
A widely cited 55.8% productivity increase from AI code completion, based on a 2023 Science paper by Doshi and Vaishnav, is narrower than commonly understood. This figure specifically measured time-to-completion for a single, well-defined task. More recent research from 2026 indicates that integrating AI tools into sustained engineering workflows over weeks yields more modest gains, around 10-20%, with significant variance across different task types like debugging or refactoring. Organizations should prioritize evaluating AI tools based on real-world, long-term workflow integration rather than isolated benchmarks. AI
IMPACT Challenges the perception of AI code completion's immediate productivity impact, suggesting more nuanced evaluation is needed for real-world workflows.