I watched a friend finish three reports in a single afternoon last month. Claude wrote the drafts, she refined them, and by 4pm she was done with what used to take a week.
So she started a fourth report. Then reorganized the team’s project tracker. Then answered emails she’d been putting off. By 9pm she was still at her desk, exhausted, wondering why she felt worse than before AI existed.
She’s not alone. And now we have the data to explain why.
The UC Berkeley Study That Changes Everything
Researchers Aruna Ranganathan and Xingqi Maggie Ye from Berkeley Haas spent eight months embedded in a 200-person tech company. Two days a week, in-person, April through December 2025. Over 40 deep interviews across engineering, product, design, and operations.
Their finding, published in Harvard Business Review: AI tools didn’t reduce work. They consistently intensified it.
Not a little. Not in edge cases. Consistently.
Here’s what they saw happening in three distinct patterns:
Task expansion. Product managers started writing code because AI made it possible. Researchers took on engineering work. Designers attempted builds they would’ve outsourced a year ago. AI lowered the skill barrier, so people absorbed tasks that weren’t theirs.
Dissolved boundaries. Workers squeezed prompts into lunch breaks, meetings, even the minutes between loading files. As HBR reported, “these actions rarely felt like doing more work… over time they produced a workday with fewer natural pauses and a more continuous involvement with work.”
Parallel processing. People ran multiple AI agents simultaneously — writing code manually while Claude generated an alternative version, reviving long-deferred tasks because AI could “handle them” in the background.
One anonymous employee put it perfectly: “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 really, you don’t work less. You just work the same amount or even more.”
By month six, reports of burnout, anxiety, and decision paralysis spiked.
It’s Not Just One Study
The Berkeley research landed in a flood of data all pointing the same direction.
Upwork surveyed 2,500 workers and found that employees reporting a 40% productivity boost from AI were also the ones burning out fastest. 88% of the highest AI-productivity workers experienced burnout — and they were twice as likely to consider quitting.
Read that again. The people getting the most value from AI are the ones closest to walking out the door.
DHR Global’s 2026 Workforce Trends Report surveyed 1,500 corporate professionals across North America, Europe, and Asia. 83% reported some degree of burnout. The top cause? Overwhelming workloads at 48%. And here’s the kicker — 52% now say burnout drags down their engagement, up from just 34% in 2025.
Wellhub polled 5,000+ employees across 10 countries: 90% experienced burnout symptoms in the past year. Nearly 40% feel them weekly. 18% feel them daily.
ManpowerGroup’s 2026 Global Talent Barometer — 13,918 workers, 19 countries — found regular AI usage jumped 13 percentage points to 45% of workers. But tech confidence fell sharply by 18%. That’s the first decline in three years.
The Jevons Paradox of Attention
Economists have a name for this. In 1865, William Stanley Jevons observed that making coal-powered engines more efficient didn’t reduce coal consumption — it increased it. Cheaper energy meant more uses for energy.
AI is doing the same thing to your attention.
When writing a report takes 20 minutes instead of 2 hours, you don’t get 100 minutes of rest. You get five more reports. When code reviews take half the time, you take on twice the codebase. When research happens instantly, every half-formed idea becomes a project worth pursuing.
Decrypt captured this dynamic well: “When workers offload rote tasks to AI, they are left with only the high-stakes decision-making and complex problem-solving… the human brain is not designed to operate at peak cognitive intensity for eight straight hours without the ‘palate cleanser’ of lower-value tasks.”
That’s the part nobody talks about. Those boring, repetitive tasks you used to hate? They were mental rest periods. And AI just eliminated them all.
Developers Are Feeling It Too
If you write code, you already know.
Index.dev reports that 84% of developers now use AI tools, and AI writes 41% of all code. But 46.4% expect burnout rates to rise. Only 21.3% think burnout will decrease.
A LeadDev survey found 22% of developers at critical burnout levels, with another 25% moderately burned out. The insight? “The exhaustion from sustained AI supervision is real but often hard to spot. The work seems easier on the surface, so developers may not recognize why they’re tired.”
You’re not writing less code. You’re reviewing more of it. You’re not solving fewer problems. You’re running more agents in parallel. And the cognitive load of checking, validating, and integrating AI output is a different kind of exhausting than writing from scratch.
The Healthcare Canary
Healthcare workers are providing an early warning.
Tebra surveyed 301 healthcare professionals and found 60% say ChatGPT has reduced burnout by streamlining documentation. Good news, right?
But then: 14% admitted emotional dependency on AI — checking it impulsively, using it to cope, feeling anxiety without access. 47% have used ChatGPT for emotional processing. And 13% said an AI tool outage would be more stressful than an electronic health records system crash.
Meanwhile, a PMC study warns that it’s “premature to assert that AI tools will reduce physician burnout.” As AI handles routine tasks, physicians are left managing only the complex, high-stress cases — all day, every day — without the cognitive breaks that simpler tasks used to provide.
What Nobody’s Saying Out Loud
TechCrunch reported that burnout hits 62% of associates and 61% of entry-level workers, but only 38% of C-suite leaders. The people making AI adoption decisions are the least affected by the consequences.
And Microsoft’s 2025 Work Trend Index found that 53% of leaders say productivity must increase — while 68% of employees already struggle with work pace and volume. The people at the top want more. The people doing the work can’t take more.
That gap is where burnout lives.
So What Do You Actually Do?
The Berkeley researchers proposed something they call an “AI practice” — structured norms around when to start, when to stop, and when to limit AI use. Think of it like digital hygiene for the AI era.
Some concrete ideas:
Protect your breaks. If you catch yourself sending prompts during lunch, that’s a signal. AI should replace tasks, not rest periods.
Set output limits, not input limits. Instead of using AI for everything you can, decide in advance how many projects you’ll take on. The tool is faster. Your brain isn’t.
Track the invisible labor. Time spent reviewing, correcting, and integrating AI output is real work. If your manager doesn’t see it, show them.
Keep some slow work. Not every task needs to be AI-optimized. Sometimes the 20-minute manual process is the mental break your afternoon needs.
Gallup’s data shows global employee engagement dropped to 21% in 2025 — the sharpest drop since COVID, costing an estimated $438 billion in lost productivity. Burning people out with AI tools doesn’t just hurt them. It costs you money.
The Bottom Line
AI doesn’t save time. It reallocates it. And without intentional boundaries, it reallocates every minute — including the ones you used to spend thinking, resting, and being human.
The researchers said it best: “Without intention, AI makes it easier to do more — but harder to stop.”
My friend? She’s started closing her laptop at 5:30. Not because she ran out of work — she’ll never run out of work again. But because she finally realized that was the point.
The question isn’t whether AI makes you more productive. It’s whether you can survive the productivity.