AI was supposed to hand workers back part of their day. Automate the repetitive pieces of a job, the reasoning went, and people would get that time back for whatever they wanted to do with it. That is not what is happening. Complaints about longer days have been circulating across tech workplaces for months, in Reddit threads, on X, in conversations between engineers comparing notes on hours that keep stretching later. A study out of UC Berkeley now backs up what those complaints describe.

Part of the reason is speed. A person now describes an app in plain language, and an AI tool writes a working version of it, a process known as vibe coding, inside a single afternoon. The output looks finished, with buttons that click and screens that load. It looks ready for customers. AI has gotten good at creating that impression fast.

Serving a handful of test users takes far less work than serving, say, 100,000 people reliably, day after day. A quick demo skips past nearly all of that difference. AI is good at producing something that runs by the afternoon. Whether that same system holds up for real customers, under real conditions, is a separate question entirely.

People outside the engineering team do not always see that distinction. When a working prototype shows up in an afternoon, it becomes easy for leadership to assume the underlying work has gotten easier too, and to cut budget or headcount on the strength of that impression. The people who remain do not see their workload shrink. They absorb whatever the cut removed, on top of the work they already had, while the output expected of the team stays the same or grows.

That kind of workload increase, doing as much or more with fewer people, is what two researchers at UC Berkeley’s Haas School of Business spent eight months examining inside a real technology company.

Xingqi Maggie Ye, a doctoral researcher at Haas, spent that time inside a 200-person technology company, sitting in on meetings, watching how employees split their day between tasks, and running more than 40 interviews across the company. She wanted to understand, in her words, how generative AI was shaping everyday work, not to prove a predetermined outcome. She and her co-author, Associate Professor Aruna Ranganathan, were checking on a simple assumption behind most corporate AI rollouts: give employees the tools, and the hours needed to finish their work should drop. Their findings, published by Harvard Business Review, show the opposite.

Employees worked faster. They took on tasks that used to belong to someone else, work that might not have been attempted at all before. They stretched their work further into the day, often without anyone asking them to. A lunch break became a moment to send one more prompt. An evening became a chance to check on a project before bed. Some kept several AI tools running at once, reviewing one document while a separate AI process worked in the background, so both the person and the software stayed in near-constant motion. The researchers call this pattern “workload creep”: each burst of extra output quietly resets what counts as a normal day, which then invites the next burst.

In the short run, that looks like a win, since a team getting more done in the same day is what companies asked the technology to deliver. Over eight months, the researchers found that pace hard to sustain. Fatigue built up. Judgment got worse. Some of the early productivity gains gave way to lower-quality work later on, the same problem a rushed prototype hides until it meets real customers. None of it added up to employees getting their day back.

That finding comes from one company over eight months, and the two researchers are still working out how far the pattern extends. What it already shows is narrower and hard to ignore: faster tools expanded what employees felt able to take on in a day, often because the work no longer had a natural stopping point, a lunch break, an evening, a weekend. The nine-to-five got fuller.

Sources

Harvard Business Review, AI Doesn’t Reduce Work—It Intensifies It – hbr.org
UC Berkeley Haas, AI promised to free up workers’ time. UC Berkeley Haas researchers found the opposite. – newsroom.haas.berkeley.edu