By rights, AI should be the great leveller, that long-promised technology that happily takes over the worst, most tedious tasks of employees, giving them much-needed breathing room to embrace meaningful work.

But while countless studies have touted AI’s capacity to cut staff time spent on the drudgery of admin or basic coding tasks, a new study published by the Harvard Business Review has found that AI isn’t reducing the burden of overwork – it’s making it a lot worse.

After an eight-month investigation into the impact of genAI tools on a midsize US tech firm, the researchers discovered that those using AI were more likely to overload themselves with work, taking on tasks outside of their role and working longer hours on a daily basis.

According to the study’s authors, the employees voluntarily took on these extra responsibilities “because AI made doing more feel possible”, a sentiment likely to elicit three cheers in the C-suite, but there is a darker side to these productivity gains.

Over a nine-month span when AI tools were embedded across the workforce, researchers observed widespread task drift, with staff stepping on each other’s toes. Researchers, for instance, took on engineering tasks, while designers and product managers began writing code, with AI filling in knowledge gaps along the way.

Although this was described by most as “just trying things” with AI, it ended with workers permanently taking on tasks that they were not trained to do, or working solo on a project that would have once taken a team.

The impact was most severe on engineers, who were forced to spend more and more time fixing AI-assisted errors made by their colleagues, finishing off half-complete code, or giving crash courses in vibe-coding to avoid catastrophe.

AI was also found to steal the day away, with workers slipping prompts into lunch breaks or during meetings, even evenings and weekends – small moments that added up to significantly less downtime and headspace.

While workers saw this as simply squeezing in a “quick last prompt,” the study found it blurred the boundaries between work and home life, with staff well‑being declining as they lost out on valuable recovery time.

Running themselves ragged, workers were also witnessed taking on several tasks at once with AI as their partner. Some would run multiple agents at once, acting as a central command, while others would delegate background tasks for the AI to handle without oversight.

This comes close to what most business leaders envision when describing the ideal AI ROI, but the human impact of this multitasking was cognitive overload, added pressure, and errors in judgment introduced through lack of attention, adding up to the threat of severe burnout.

Recommended reading

• 79% of Firms Betting On AI To Raise Revenue, Only 1 in 4 See How
• AI is Rendering The Way We Measure Productivity Obsolete
• Workers Spend 3 Hours A Week Dealing with ‘AI Workslop’

But with businesses laser-focused on increasing their productivity gains, AI will continue to be more tightly woven into everyday workflows.

The study’s authors argue, however, that rather than keep a foot on the gas, employers must realise that AI is only as effective as the human at the helm, and take steps to safeguard their staff.

That means encouraging intentional pauses to regulate the flow of work; batching or sequencing non‑urgent AI outputs to reduce fragmentation; and protecting time for human connection to restore perspective.

“The promise of generative AI lies not only in what it can do for work, but in how thoughtfully it is integrated into the daily rhythm,” the authors conclude.

“Our findings suggest that without intention, AI makes it easier to do more—but harder to stop.”