Many predict AI will replace much of what people do, but there is also a view that this is not the case, at least for now. A recent survey by enterprise software company Workday shows people are still spending a lot of time dealing with problems from using AI.

From employees’ perspective, AI has sped up work, but it has also created many other tasks.

Axios reported that slogans promising productivity gains from AI are pouring out, but in reality scenes are emerging in which AI leads people to do more work, not less.

Workday surveyed 3,200 corporate users of AI in North America, Europe and Asia. Half of them are in leadership roles. The respondents work at companies with at least $100 million in revenue and at least 150 employees. The survey did not mention which AI products they use or which companies developed them.

The survey found 85 percent of respondents said AI saved them about 1 to 7 hours a week. But 37 percent of the time saved was spent on so-called rework such as fixing, rewriting and verifying results generated by AI. Only 14 percent of respondents said they consistently get positive results from AI.

Gerrit Kazmaier, Workday’s president of product, told Axios about the survey, "There is a big productivity paradox." He said, "The people who use AI most often are the ones spending the most time reviewing and fixing what AI produces."

The Workday survey is not an outlier. Research from MIT and Harvard Business Review (HBR) that questions AI’s productivity gains is in a similar vein.

In September last year, HBR said AI should make work easier but is instead creating a new problem called "workslop." HBR described workslop as low-quality AI-generated content that leads employees to waste time on tasks such as writing memos, reports and emails.

HBR researchers surveyed 1,150 U.S. adults who identified themselves as office workers in August and September about their experience with workslop. The researchers did not name the term and only provided a definition. The survey found 40 percent of respondents said they experienced situations in the past month in which AI slowed their work. Each case took an average of 1 hour and 56 minutes. Axios, in covering the Workday survey, said there are reasons the term workslop has become popular.

Axios cited consulting firm AlixPartners co-CEO Rom Horneby as saying CEOs and employers are eager to maximise productivity benefits from AI, especially to cut labour costs, but for now AI is being used as a pretext to justify layoffs.

A survey of CEOs by AlixPartners found 95 percent of respondents said they will carry out layoffs within 5 years due to AI. On this, Horneby told Axios, "This is closer to hope than reality. CEOs are not yet seeing productivity improvements from AI."

Survey findings keep coming out showing it is still difficult to feel productivity gains from AI, and that can be seen as a backdrop to the AI bubble theory continuing to have traction.

The continued questions over AI-driven productivity gains may be a sign of an AI bubble, but many also see it as trial and error as a new technology takes root, like the internet. Cowork, recently introduced by Anthropic to automate repetitive office tasks, is also cited as one example showing AI capabilities are evolving quickly. Cowork is said to have taken less than 1.5 weeks to develop thanks to AI.

Outside coding, such claims seem to be heard relatively less often. While the growth of AI coding continues despite concerns that code written with AI coding tools could be vulnerable from a security perspective, it remains to be seen whether this year will bring coding-level use cases in other areas of work.