Seventy-seven percent of employees say AI has increased their workload. Not decreased. Increased.
That number comes from an Upwork survey of 2,500 workers, and it lines up with a growing pile of research all pointing in the same direction: the promise that AI would free us from busywork is not playing out the way the pitch decks said it would.
A Wall Street Journal report this week pulls together what has become hard to ignore. AI tools speed up individual tasks, but total work volume keeps growing because expectations ratchet up to match.
The Berkeley Study That Explains the Mechanism
Researchers at UC Berkeley spent eight months embedded in a 200-person tech company, observing how AI tools changed daily work. Published in Harvard Business Review, the study identified three specific ways AI makes work heavier:
Task expansion. Product managers and designers started writing code because AI made it feel accessible. Engineers then spent more time reviewing and fixing that AI-assisted code from non-engineers. The work didn't disappear. It shifted and multiplied.
Blurred boundaries. Workers described sending "one quick last prompt" before leaving their desks. AI's always-available, conversational interface bled work into lunch breaks, meetings, and evenings. As one engineer put it: "You had thought that maybe you could work less. But then really, you don't work less. You just work the same amount or even more."
Forced multitasking. People ran multiple AI threads simultaneously while doing manual work, fragmenting attention across more tasks than before.
The Numbers Keep Getting Worse
Boston Consulting Group surveyed 1,488 U.S. workers in March 2026 and found a tipping point: productivity peaked at three or fewer AI tools. Once workers juggled four or more, self-reported productivity dropped. Workers using AI heavily reported 14% more mental effort, 12% greater fatigue, and 19% more information overload. BCG calls this "AI brain fry," and 34% of those experiencing it are actively looking for new jobs.
A Workday survey of 3,200 employees found that nearly 40% of AI-related work time goes to checking and fixing AI output. Only 14% of workers see consistently positive outcomes.
Even developer productivity data tells this story. One study found that developers perceived a 20% speed boost from AI coding tools, but the actual time to complete work was 19% longer than expected. The output felt faster. The calendar said otherwise.
Perhaps the most striking gap is between executives and everyone else. An NBER survey of nearly 6,000 CEOs and CFOs found that 90% of firms using AI reported zero measurable productivity or employment impact. Meanwhile, 98% of bosses believed AI was saving their employees time, while 40% of those employees said it saved them nothing.
What This Means for People Buying AI Tools
Goldman Sachs found "no meaningful relationship between productivity and AI adoption" at the economy level, with only customer service and software development showing clear gains.
This does not mean AI tools are useless. It means most organizations are deploying them without changing how work is structured. They add AI on top of existing processes, then raise output expectations because "you have AI now." The tool gets faster. The job gets bigger.
For anyone evaluating AI tools for their team or their own workflow, the research suggests a practical ceiling: stick to two or three tools you actually know well. The returns from adding a fourth, fifth, or sixth appear to be negative. And build in explicit boundaries. AI's low-friction interface makes it easy to work during every gap in your day, which is not a feature if it leads to burnout.
The 96% of C-suite leaders who expect AI to boost productivity are not wrong about the technology's capability. They are wrong about what happens when you hand it to an already-overloaded workforce without changing anything else.