Why AI makes you work more, not less, according to Berkeley

You adopted every AI tool your company threw at you. You automate emails, generate reports in seconds, and let copilots write your code. You're completing 40% more tasks than last year. And you've never been more exhausted.

A February 2026 study from UC Berkeley's Haas School of Business, published in Harvard Business Review, tracked 200 employees at a U.S. tech company for eight months. Researchers Aruna Ranganathan and Xingqi Maggie Ye discovered something that should make every productivity enthusiast uncomfortable: the workers who embraced AI most aggressively didn't work less. They worked the same amount or more, while their cognitive load quietly spiraled.

The popular advice is wrong

The standard pitch goes like this: adopt AI, save hours, reinvest that time into "high-value work." Every vendor deck, every LinkedIn thought leader, every corporate memo repeats it. And it sounds right because, on paper, AI genuinely does make individual tasks faster.

But here's what nobody mentions. When the cost of starting a task drops to near zero, the number of tasks you attempt doesn't stay the same. It explodes. The Berkeley researchers documented three specific mechanisms driving this silent overload.

Task expansion: you're doing jobs that aren't yours

Product managers started writing code. Engineers began handling design reviews. Researchers took on operational work they'd previously outsourced. Because AI made it possible to do more, employees voluntarily absorbed responsibilities outside their roles, without anyone asking them to.

One engineer in the study put it bluntly: "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."

If you've noticed your to-do list growing despite all the "time-saving" tools, this is why.

The invisible bleed: work without edges

The second mechanism is subtler and arguably more dangerous. Prompting an AI doesn't feel like work. It feels like a conversation. So employees started sending "one last prompt" during lunch, after dinner, on weekends. The Microsoft 2025 Work Trend Index found that knowledge workers now send 58 chats daily outside work hours (a 15% year-over-year increase) and get interrupted every two minutes during core hours, up to 275 times per day.

The result: 68% of employees now struggle with work pace and volume, and 46% report outright burnout. Not because they're lazy or disorganized, but because AI erased the natural friction that used to create boundaries.

Why 3 companies cracked it (and what they did differently)

The Berkeley researchers identified a framework they call "AI Practice," and a handful of organizations have already operationalized it. The approach has three components that work together, not individually.

1. Capacity caps, not productivity goals. Instead of measuring output ("complete more tasks"), these teams set maximum cognitive load thresholds. When a team hits capacity, new work gets queued, not piled on. This directly counters the task expansion problem. Microsoft's own research found that Frontier Firms using this approach report 71% of employees thriving, compared to just 37% at typical organizations.

2. Sequenced attention blocks. Rather than letting AI enable constant multitasking, these companies batch AI interactions into dedicated windows. Notifications get clustered. The "one last prompt" impulse gets structural guardrails. The goal isn't to use AI less; it's to use it in focused bursts rather than an ambient drip.

3. Human grounding rituals. The most surprising finding: teams that scheduled brief, non-AI dialogue (structured check-ins about decisions, not status updates) maintained sharper judgment. The Berkeley study found that employees who replaced interpersonal discussion with AI consultation made faster but measurably worse decisions over time.

The math your manager hasn't done

Here's the calculation that changes everything. The National Bureau of Economic Research found that average time savings from AI adoption amount to roughly 3% of work hours. Meanwhile, Microsoft's data shows 60% of the workday is already consumed by communication overhead. You're optimizing 3% while hemorrhaging 60%.

The companies that solved this didn't start by adding AI. They started by subtracting: removing unnecessary communication loops, eliminating task sprawl, capping workload before it hit the burnout threshold. Then they added AI into a system designed for human cognitive limits.

What to do on Monday morning

Stop measuring your productivity by tasks completed. Start measuring it by decisions made well. Pick your three highest-leverage tasks for the day and batch your AI usage around them. Block one hour with zero digital interrupts. And the next time someone suggests you "just use AI for that too," ask yourself: is this expanding my capability, or just expanding my workload?

The 200 employees in that Berkeley study weren't failing. They were succeeding themselves into exhaustion. The question isn't whether AI makes you more productive. It's whether anyone set a ceiling before the floor gave out.