Does AI Make You Work Less? A Berkeley Study Found the Opposite

No, not on its own. In the closest study so far, the people who used AI the most did more work, not less. UC Berkeley researchers spent eight months inside a 200-person tech company and ran more than 40 interviews. Nobody was told to hit new targets. People took on more because the tools made more feel doable, and work spread into lunch breaks and late evenings (TechCrunch, 2026). The fix for an operator: hand AI a whole job with a finish line, then check that your hours moved.

What did the Berkeley study find?

That AI intensifies work rather than reducing it. The research, published in Harvard Business Review on February 9, 2026, is by Aruna Ranganathan, an associate professor at UC Berkeley's Haas School of Business, and Xingqi Maggie Ye (Harvard Business Review, 2026). Inside the 200-person company they studied, employees worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day, often without being asked to (Harvard Business Review, 2026).

To-do lists expanded to fill every hour AI freed up, then kept going. The authors call the result workload creep, cognitive fatigue, burnout, and weakened decision-making (Harvard Business Review, 2026). One engineer put it plainly: "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" (TechCrunch, 2026).

One caution: it is one company and in-progress research, and the full HBR article is paywalled, so the details above come from the authors' summary and TechCrunch's report. The authors' fix is an AI practice: norms that include intentional pauses, sequencing work, and more human grounding (Harvard Business Review, 2026). For a small business that is one question: did the task leave your plate, or did you just do it faster?

Are the productivity gains even real?

Smaller than they feel, where anyone measured. In a randomized trial METR published in July 2025, 16 experienced open-source developers worked 246 real issues in their own codebases. With AI allowed, they took 19% longer. Going in, they expected AI to make them 24% faster. Afterward, they still believed it had made them 20% faster (METR, 2025).

METR has since said the picture has moved. In a February 2026 update on 57 developers, it wrote that developers are likely more sped up by AI now than in early 2025, adding that "because of the selection effects in our experiment, our data is only very weak evidence for the size of this increase" (METR, 2026).

Outside software, hours did not move either. Economists Anders Humlum and Emilie Vestergaard linked surveys of 25,000 workers across 7,000 Danish workplaces to government labor records. Most employers had adopted chatbots and workers reported gains, yet the authors estimate "precise null effects on earnings and recorded hours," ruling out effects larger than 2% two years after ChatGPT launched (National Bureau of Economic Research, 2025). An earlier version of the paper, cited by TechCrunch, put time savings at about 3% (TechCrunch, 2026). What changed was the work itself: new tasks in content generation, AI oversight, and AI integration (National Bureau of Economic Research, 2025). The AI did not take work off the table. It changed the work and added supervision.

Why does AI add work instead of removing it?

Because a faster tool and a hand-off are different things. When AI makes you quicker at a task you still own, the finish line moves. The proposal gets a second draft. The reply goes out tonight. Nobody at the Berkeley company set that pace from above. Over eight months, people set it on themselves (TechCrunch, 2026).

Service business owners already live at that ceiling. In Xero's February 2026 survey of 300 US small business owners, respondents spend about eight hours a week consumed by worry, 74% said stress had hurt their work, and 40% had considered giving up the business (Xero, 2026). Xero sells accounting software and commissioned the survey, so read it as a vendor survey with a stated sample. An owner's day is already full. Make an owner faster at answering texts and the texts get answered faster. The day stays the same length.

Hours come back only when a task leaves your plate with its outcome attached. A front-desk hire does that: the phone gets answered whether or not you are in the building. An AI employee has to clear the same bar. Our guide to AI employee or new hire prices the two side by side.

How does an operator keep the hours AI saves?

Give AI whole tasks with a clear finish line, then measure the outcome instead of your speed. The Berkeley researchers watched a 200-person company feel faster and work longer at the same time (TechCrunch, 2026), so the test is not how you feel. Three questions separate a hand-off from a speed-up.

Does the task end without you? Answering a call, booking the slot, sending the reminder, requesting the review: each has a done state someone else could check. Drafting your own proposal does not. You still edit and send it.

Is the outcome written down? "Every call answered within three rings, every booking on the calendar" is a job. "Help with the phones" is a tool. Our guide to what an AI receptionist can handle lists the jobs that pass this test.

Did your hours actually move? Log your week before the change and again 30 days after. If the AI took the phones and you still work the same hours, you refilled the time.

Worked example: A dental office (an illustration, not a customer) spends 15 front-desk hours a week on calls, reminders, and reschedules (assumption). Option A: give the coordinator an AI writing assistant. Calls still ring to the desk, reminders go out faster, and 15 hours become 13 (assumption). The two freed hours fill with the insurance backlog by Wednesday. Option B: an AI employee answers overflow and after-hours calls, sends every reminder, and books reschedules into the calendar. Phone time drops to the calls that need a person.

Only option B moves hours, and the office can prove it with two logged numbers: calls answered and appointments kept. Even then, the hours stay saved only if the office decided in advance what they were for.

What to do this week

1. List the tasks with a finish line. Every repeat task another person could check as done: calls answered, appointments booked, reminders sent, reviews requested. Those are hand-off candidates.
2. Baseline your hours. For one week, note when you and your front desk start and stop, and how many hours go to the phone and follow-up. Without this number you cannot tell whether the AI worked.
3. Hand off one whole job. Start with the task that costs most when it slips. For most service businesses that is the unanswered call; price yours with the missed call cost calculator. Include the rule for when a person takes over.
4. Set the response standard once. Decide what "fast enough" means for a new lead and write it down (our speed to lead guide has the numbers). A fixed standard stops expectations from ratcheting up every month.
5. Re-measure at 30 days. Compare hours, calls answered, and appointments kept against the baseline. If hours did not fall, decide what the freed time is for and block it on the calendar.

Where Nirvani fits

Nirvani captures incoming calls, texts, social messages, and website chat in one system. Its AI employees answer, qualify leads, book appointments into your calendar, handle basic scheduling, follow up on no-shows, and send review requests, 24/7, in your business's voice, around the processes your team already uses. Each is a whole job with a finish line. N1 is the entry operating system. See how the AI employees work.