If you've read any tech investment memo in the past year, you've likely seen the same narrative: AI dramatically boosts productivity, letting employees do less work while producing more. A nine-month study from Berkeley's Haas School of Business challenges that claim with hard data.
Economists Aruna Ranganathan and Xingqi Maggie Ye published their findings in Harvard Business Review. From April to December 2025, they tracked 200 employees at a U.S. tech company, observing behavioral changes after AI tools were introduced. The conclusion is blunt: AI did not reduce work; AI intensified work.
New 'Multi-Threading' Work Rhythm: Efficient on Surface, Draining in Reality
The research team found that AI tools didn't actually reduce task volume or working hours. Instead, they created a novel work rhythm — 'multi-threading.' Employees wrote code manually while letting AI generate alternative versions, ran multiple AI agents for different tasks simultaneously, and even resurrected long-dormant projects because 'AI can handle them in the background.'
On the surface, it looked like productivity gains: each person juggled more projects and output speed increased. But researchers observed real costs: constant attention switching, frequent checking of AI outputs, and a growing to-do list, all leading to high cognitive load and a feeling of 'always juggling.' Employees did get more done, but they felt drained after work.
Developer Confession: Burnout in One or Two Hours
Prominent developer Simon Willison reposted the study on his blog and admitted his own experience matched closely. As one of the most active Large Language Model (LLM) practitioners globally, Willison regularly shares his AI workflows. He said he can indeed push two or three parallel projects simultaneously, but the price is mental exhaustion within one to two hours. He also observed similar patterns among other developers: one colleague kept writing till 3 a.m. due to 'just one more prompt,' severely disrupting sleep quality.
The Real Problem: Lack of Organizational Norms for AI Use
The key insight from the research team lies in their diagnosis of the cause: organizations lack structural norms for AI use. Most companies simply buy licenses, hand out accounts, attach a PDF of 'best practices,' and expect employees to figure it out on their own. The researchers recommend that companies establish formal AI practice frameworks, clearly defining where AI should and should not be applied, and distinguishing 'real efficiency gains' from 'just pedaling harder in place.'
A Stress Test for the Valuation Logic
Over the past year, 'AI boosts productivity' has been a core logic driving tech stock valuations. From Nvidia to Microsoft, from OpenAI to AI agent startups, the entire industry chain's valuation premise is that AI will make knowledge workers 2-to-10 times more productive. But if Berkeley's research is correct — that AI's actual effect is not 'doing less' but 'doing more and getting more tired' — that valuation logic needs recalibration. Productivity improvement and work intensification are two different things. The former cuts costs and boosts margins; the latter boosts output short-term but risks burnout, higher turnover, and quality decline long-term.
None of this means AI has no value. But its value may not be about 'letting people do less.' It may be about 'letting people do different things.' And different doesn't automatically mean easier.
Adjustment Period: Decades of Work Habits Upended in Two Years
The explosive adoption of AI from 2023 to 2025 essentially demanded that the entire knowledge economy relearn how to work in just two years. This relearning won't happen automatically; it requires deliberate organizational design, managerial mindset updates, and most importantly, acknowledging that 'more output' and 'better work' are two entirely different things. Silicon Valley loves the term '10x engineer.' AI promised to turn everyone into a 10x. But this research suggests we might not be getting 10x efficiency — we might be getting 10x fatigue.