The first signs of burnout are coming from the people who embrace AI the most

In today’s American workplace, the most compelling storyline is no longer that artificial intelligence will replace workers — it’s that it will rescue them from overwhelming workloads.

That message has been promoted heavily over the past three years to professionals uneasy about automation. While some white-collar positions may vanish, the prevailing pitch suggests that for many roles, AI acts as a performance enhancer. Lawyers, consultants, writers, programmers, analysts, and others are told they can become more efficient and more valuable. The promise is straightforward: the tools handle part of the load, the individual works more effectively rather than harder, and productivity increases without added strain.

However, a recent study featured in Harvard Business Review takes that optimistic narrative to its logical conclusion—and the outcome looks less like liberation and more like exhaustion. Instead of a sweeping productivity breakthrough, the findings suggest that organisations could unintentionally become engines of burnout.

The research, described as ongoing, involved UC Berkeley scholars spending eight months embedded in a 200-person technology firm. Their goal was to observe what actually happens when employees fully adopt AI tools. Through more than 40 detailed interviews, researchers discovered that workers were not formally pushed to meet higher quotas or performance benchmarks. No directives demanded increased output. Yet employees naturally began expanding their workload simply because AI made additional tasks feel achievable. As new efficiencies opened up time during the day, that time quickly filled. Work crept into lunch periods and stretched into evenings. Task lists grew to occupy the hours automation initially freed—and then extended beyond them.

One engineer captured the experience succinctly: “You had thought that maybe, oh, because you could be more productive with AI, then you save some time, you can work less. But then, you don’t actually work less. You work the same amount or even more.”

A similar sentiment surfaced on the technology discussion platform Hacker News. A commenter wrote, “I feel this. Since my team has jumped into an AI everything working style, expectations have tripled, stress has tripled, and actual productivity has only gone up by maybe 10%. It feels like leadership is putting immense pressure on everyone to prove their investment in AI is worth it, and we all feel the pressure to try to show them it is while actually having to work longer hours to do so.”

The situation is both intriguing and concerning. Conversations about AI’s role in employment often stall on a single debate: whether the productivity gains are meaningful. Far fewer discussions consider what would happen if those gains materialise.

The new findings do not stand entirely on their own. A separate experiment conducted last summer indicated that experienced software developers who used AI tools actually required 19% more time to complete certain tasks, even though they believed they were working 20% faster. Around the same period, research from the National Bureau of Economic Research, which examined AI implementation across thousands of workplaces, found that average productivity improvements translated to only about 3% in time savings, with no notable effects on wages or working hours across occupations. Both sets of results have been scrutinised and debated.

What may make the latest study more difficult to dismiss is that it does not dispute AI’s ability to enhance individual performance. Instead, it acknowledges that augmentation can occur — and then highlights the consequences. According to the researchers, greater capacity often results in “fatigue, burnout, and a growing sense that work is harder to step away from, especially as organisational expectations for speed and responsiveness rise.”

The technology sector has largely assumed that enabling people to accomplish more would solve longstanding workplace challenges. Yet this research suggests that amplifying output without redefining boundaries may introduce an entirely new set of problems.