The pitch has always been the same: AI handles the tedious stuff, you get more time for the meaningful work. Less grinding, more thinking. The tools arrive, productivity goes up, and somewhere in the calendar a few hours open up.
That’s not what’s happening. Two new studies — both published in Harvard Business Review within the past month — paint a much more complicated picture of what AI is actually doing to people at work. And one of them introduces a term that might already sound familiar: AI brain fry.
Faster, Broader, Longer
The first study, out of UC Berkeley’s Haas School of Business, tracked 200 employees at a US tech company over eight months. Researchers conducted over 40 in-depth interviews across engineering, product, design, and operations and found that AI tools didn’t reduce workloads — they expanded them.
Workers using AI completed more tasks, tackled a wider variety of responsibilities, and extended work further into their days, often without being asked. Having a capable AI “partner” created momentum — and that momentum was hard to stop. People started reviving long-deferred projects because AI could “handle them” in the background. They ran multiple agents in parallel. They prompted AI tools during lunch breaks and weekends. The boundary between work hours and off hours quietly dissolved.
What looked like a productivity revolution turned out to be unsustainable work intensification, with cognitive fatigue and burnout starting to offset any measurable gains.
A Buzzing in the Brain
The second study goes further — and gets more specific about what this overload actually feels like. Researchers at Boston Consulting Group surveyed more than 1,400 full-time US workers at large companies and found 14% had experienced “AI brain fry”: mental fatigue caused by excessive use or oversight of AI tools beyond a person’s cognitive capacity.
Workers described the condition using remarkably consistent language. A “buzzing” sensation in the head. Mental fog. Difficulty concentrating. Slower decision-making. Headaches. The symptoms were most common in marketing, software development, HR, finance, and IT roles — essentially any position now expected to manage AI outputs as part of a normal workday.
Brain fry wasn’t just uncomfortable — it had measurable consequences. Workers experiencing it reported more frequent errors, including major mistakes with real consequences, and were more likely to say they were thinking about leaving their jobs.
The Oversight Trap
The single most draining pattern researchers identified wasn’t using AI — it was watching it. Employees who spent significant time monitoring AI outputs reported 12% more mental fatigue than those who didn’t. Managing clusters of agents generating code, drafts, and summaries at high speed means tracking more outcomes across more tools within the same number of hours. The workload doesn’t shrink; it just changes shape.
One AI founder described it this way online: ending each day exhausted — not from the work itself, but from the managing of the work. That distinction is important. The fatigue isn’t about effort. It’s about the relentless cognitive load of supervising systems that never slow down.
Where AI Actually Helps
The research isn’t entirely bleak. Both studies found that AI use genuinely reduced burnout in one clear scenario: when it replaced routine and repetitive tasks. Offloading genuinely boring work — the kind that drains attention without producing anything interesting — created space for creative thinking and higher-value tasks.
The problem is that most organizations aren’t deploying AI that way. They’re deploying it in ways that maximize output metrics, which means workers end up juggling more tools, tracking more outcomes, and producing more — right up until they can’t.
More Output, Less Capacity to Think
The Berkeley researchers suggest organizations need to build an intentional “AI practice” — structured norms around how tools get used, including built-in pauses and space for human connection — to stop short-term productivity gains from becoming a long-term burnout pipeline.
That’s a reasonable prescription for companies. For individuals already deep in AI-assisted workflows, the more immediate question might be simpler: if the tools are making you feel worse, faster isn’t actually better.