According to a study by the Work AI Institute, which surveyed 6,000 office workers in the U.S., the U.K., and Australia, employees save an average of about 11 hours per week thanks to AI tools. However, more than half of this time savings—over 6 hours—is spent verifying, correcting, and reviewing the results generated by algorithms.
In fact, it’s not just about speeding up work, but also about redistributing the workload: instead of performing routine tasks, employees take on the roles of “supervisors” and editors of AI-generated work.
The survey shows that 75% of respondents do indeed feel an increase in personal productivity. But at the company level, the effect is noticeably weaker: only 13% of organizations report tangible business growth after implementing such technologies.
According to Professor Paul Leonardy of the University of California, companies underestimate the amount of hidden work that arises around AI tools. This involves not only formulating queries and preparing context, but also constantly verifying results, searching for errors, and manually refining data.
According to the study, the breakdown of time usage looks like this: about 37% is spent interacting with AI (queries, clarifications, iterations), another 36% on applying the results to real-world tasks, and the rest on verification and corrections.
A practical case study is also telling: a junior developer implemented thousands of lines of AI-generated code into a project, after which the system stopped working. A senior engineer had to figure out the error—the author of the changes himself could not explain the logic behind the code.
The study’s conclusion is phrased cautiously: AI does indeed speed up certain stages of work, but at the same time creates a new layer of tasks—management, verification, and “manual stabilization” of the results it generates.
