Artificial intelligence is poised to transform office work, yet executives and employees report underwhelming productivity gains. Despite 69% of firms actively using AI, nearly 90% of surveyed executives say the technology has impacted neither employment nor productivity. Researchers from the Federal Reserve Bank of Atlanta, Stanford University, the Bank of England, and the Autonomous Technological Institute of Mexico attribute this to a "verification tax." This phenomenon describes the human effort required to check AI output for accuracy and safety, creating a bottleneck that offsets potential efficiency.
MIT Sloan's Christian Catalini emphasizes that human verification bandwidth is the current binding constraint on growth. Surveys indicate workers spend significant saved time correcting low-quality AI output. A Workday poll of 3,200 found employees use 37% of saved time for corrections.
Google economists found scientists spend nearly 90% of saved time checking AI outputs, with 46% spending over a quarter of that time. This rework suggests AI generates solutions quickly, but physical testing remains a slow downstream bottleneck. MIT and Wharton research on 500,000 GitHub developers shows AI tools increase activity, but gains shrink as work shifts from writing code to shipping software.
The constraint is shifting from coding to human review, integration, testing, and release. While AI saves time, automating a task often begins a new workflow phase requiring human oversight, particularly in tech where code review is essential for total efficiency.
Source: New York Times Business · Summarized by HeadlinesBriefing