Adopting generative AI (genAI) tools helps employees work faster, take on more tasks and extend their workdays, but new research suggests the changes may be unsustainable, leaving workers doing more overall and risking burnout.

Researchers at the University of California Berkeley's Hass School of Business spent eight months observing work habits at a US technology company with around 200 employees.

They regularly observed and interviewed engineering, product, design, research, and operations staff as they voluntarily began using genAI in their jobs.

Workers found the tools empowering. AI gave them confidence to attempt tasks outside their normal roles, creating what researchers described as an “intensification” of work. Product managers, for example, began writing code, while researchers took on engineering tasks, often guided by AI feedback rather than colleagues.

This made workers feel empowered as they tried “newly accessible” projects with the support and feedback not of peers, but of AI: for example, product managers began writing code, while researchers undertook engineering tasks.

Many employees tapped AI to do more work in the same time – for example, filling what used to be work breaks with a series of AI prompts to do a range of small tasks – while others spent extra time, like engineers helping colleagues learn vibe coding.

New small projects, “just trying things” with genAI and extra work responsibilities “accumulated into a meaningful widening of job scope,” wrote authors Xinqi Maggie Ye and Associate Professor Aruna Ranganathan of UC Berkeley’s Hass School of Business.

“These actions rarely felt like doing more work, yet over time they produced a workday with fewer natural pauses and a more continuous involvement with work,” they found, citing “a new rhythm in which workers managed several active threads at once.”

AI’s efficiency paradox laid bare

GenAI is often promoted as a way to boost productivity and make life easier for workers by automating menial work like coding, giving them more time for high-level thinking, but the observational study’s findings suggest that real outcomes are far different.

“Once the excitement of experimenting fades,” they found, “workers find their workload has quietly grown and feel stretched from juggling everything that’s suddenly on their plate.”

The results corroborate similar studies around the world that have found despite the rhetoric, the outcomes of AI adoption still vary widely and the evidence for its benefits, as the CSIRO puts it, are “murky”.

While companies like Procter & Gamble and Boston Consulting Group reported productivity gains after adopting AI – as did one PwC analysis, one study of 300 Australian employees found 30 per cent did not see productivity benefits.

A recent Workday survey of 3,200 employees found that despite saving one to seven hours per week thanks to AI, employees were also spending nearly 40 per cent of their time checking its output and fixing mistakes.

Just 14 per cent of those workers said they “consistently get clear, positive net outcomes from AI”, with 77 per cent admitting they spend as much time reviewing AI-generated work as they do work done by humans, or even longer.

One observer calls this a “rework crisis no one’s measuring”, with developers in particular reporting that coding with AI tools feels 20 per cent faster but actually takes 19 per cent longer because AI-generated ‘workslop’ just isn’t ready for production.

Executives are blind to workers’ lived AI experience

Misapprehensions about genAI’s benefits have distracted executives for whom the technology has proven to be a lingering blind spot, with UNSW Business Schools Frederik Anseel warning businesses to stop focusing on “easily measured outputs”.

“Employees learn to game these systems, focusing on appearing busy rather than generating genuine value,” he wrote.

By all accounts, executives are buying it: AI consultancy Section’s recent AI Proficiency Report, for one, surveyed 5,000 US, UK and Canada knowledge workers and warned that “executives are in the dark” about how genAI is helping them and their workers.

Some 69 per cent of workers are ‘AI experimenters’ using genAI for basic tasks, while 28 per cent are still ‘AI novices’ – with 97 per cent of workers using AI poorly or not at all, 25 per cent saying they save no time with AI, and 40 per cent happy to never use AI again.

“Employees know how to use an LLM, but bounce off when they can’t think of a use case for it,” the report notes, with less than a third of respondents reporting that they save four hours or more per week by using AI.

“Executives said their company has a clear AI strategy, adoption is widespread, and that employees are encouraged to experiment and build their own solutions,” Section COO Taylor Malmsheimer observed, but “the rest of the workforce disagrees.”