Where’s the holistic AI productivity data?

This article critically examines the rush to adopt AI tools in companies without measuring actual costs and productivity gains. The author, drawing from experience running a small company, argues that while AI can boost individual productivity for skilled users who can review and correct outputs, these gains are limited and often overestimated. The piece highlights that successful AI use requires deep domain knowledge to spot errors, and that agentic coding often produces only superficial results. The author calls for more rigorous, holistic research on AI's true impact on productivity, noting current data focuses on narrow tasks rather than overall business outcomes.