Despite loud promises, artificial intelligence has so far been less about freeing people from routine work than about shifting onto them the task of fixing its mistakes. According to a survey by Workday published on Wednesday, employees continue to spend a significant share of their working time dealing with the fallout from AI output.
The core contradiction is that AI simultaneously speeds up tasks and expands their volume. Research conducted by Workday last November shows that 85% of surveyed employees believe AI saves them between one and seven hours a week. However, about 37% of those time savings are consumed by so-called “rework”—correcting errors, rewriting text, and verifying results. Only 14% of respondents said they consistently obtain positive outcomes from AI.
The survey included 3,200 employees who use AI, half of whom hold managerial positions. All work at companies in North America, Europe, and Asia with revenues of at least $100 million and staff counts of 150 or more. The report does not specify which AI products participants used or which companies developed them.
“We are seeing a serious productivity paradox,” said Gerrit Kazmaier, Workday’s president of product. According to him, the most intensive users of AI also spend the most time checking and correcting what it produces. These findings echo other studies, including work by MIT and Harvard Business Review, which likewise question the reality of productivity gains. It is no coincidence that the term “workslop” has entered common use.
The problem, Kazmaier argues, is that familiar logic no longer holds. Normally, the better a person masters a technology, the more efficient they become. With AI, rising competence instead leads to a clearer understanding of where and how the system goes wrong. He cites users who run the same prompts across multiple AI models and then cross-check the results against one another.
At the business level, expectations are even more inflated. Company executives assume that AI will boost productivity and reduce staffing costs.
In practice, however, says Rob Hornby, co-CEO of the consulting firm AlixPartners, AI more often serves as a convenient explanation for layoffs whose causes lie elsewhere. In a survey by his company, also published on Wednesday, 95% of chief executives said they expect job cuts over the next five years because of AI. In Hornby’s view, these are hopes rather than a reflection of reality: managers have yet to see tangible productivity gains.
At the same time, he concedes that in certain niches AI does deliver results—for example, in some forms of low-level, standardized writing. But overall, he says, “it is very difficult for us to prove the existence of real productivity benefits.”
The situation may change as the technology matures. AI tools are improving rapidly: a new Anthropic product designed to automate office routine, for example, was built in less than a week and a half, with all of its code written by AI itself. Even so, the adoption of any new technology takes time. Employees must learn how to use it, companies must integrate it into their processes, and developers must create solutions that genuinely deliver value. It is at this stage that Workday and other enterprise software providers are trying to sell their AI products.
The result is a paradox: AI speeds up work, yet quietly adds new tasks along the way.