Length of the on-site study: 8 months (Source: UC Berkeley Haas researchers Ye and Ranganathan, as reported by InfoWorld (September 29, 2026))
Saved time gets spent
Many AI business cases assume faster work leaves people with spare hours. A field study from UC Berkeley Haas suggests those hours may not appear. The time is spent on more work.
This suggests that AI speed is not free capacity. It is a choice about what to take on next. If leaders do not make that choice, the workload makes it for them.
What the researchers watched
Doctoral researcher Xingqi Maggie Ye and Associate Professor Aruna Ranganathan studied a 200-person technology company where staff had broad access to generative AI tools. Ye spent eight months on site. She watched how people structured their days, attended meetings and ran more than 40 interviews.
The work is in progress and was featured in Harvard Business Review. It covers one company. It does not measure how common the pattern is elsewhere.
Three ways the work grew
The researchers say employees worked faster, took on wider tasks and stretched work into more hours, "often without being asked to do so." Ye described three forms.
First, people took on work that once belonged to someone else, or that nobody would have attempted. Second, work crept into pauses. People sent prompts at lunch, before meetings and in the evening. Third, people kept many threads alive at once. Some ran several AI agents at the same time.
Ye said the effect can feel good in the moment. Yet when people looked back on their overall experience, some described feeling busier, more stretched and less able to disconnect.
How a win turns into a baseline
Ye warns of a cycle. More capability produces more output. More output raises expectations. Higher expectations push people to take on still more. What was extra effort becomes standard performance.
She adds that constant switching and less recovery can impair judgment and raise errors. The result is that organisations may struggle to tell real productivity gains from unsustainable intensity.
Think of a road widened to ease traffic. Drivers fill the new lane. Faster coding may work the same way, with the road getting busier instead of clearer.
The engineering version
InfoWorld's coverage applies the idea to software teams. It quotes developer Simon Willison, who says coding agents make software engineering harder because using them well takes "extraordinary discipline and knowledge." InfoWorld notes he is a fan of the tools and was not criticising them.
Geoffrey Huntley called the work "incredibly taxing." He reported roughly 16-hour days over 12 days, and said this was his choice, not a work requirement. InfoWorld stresses that such comments are not representative measurements. They come from enthusiastic users.
InfoWorld offers an example. A team uses faster coding to finally attempt a delayed migration. The engineers may write less code. They may spend more of the day on compatibility, customer disruption and which old behaviours must survive.
What leaders should ask
InfoWorld cautions against budgeting as if AI has both enlarged the road map and reduced the need for engineering expertise. Both cannot be assumed.
Ye's answer is what she calls an "AI practice." It means short, structured pauses before major decisions, batching non-urgent updates, protected focus windows, and time for check-ins and shared reflection. She says the aim is not to slow innovation.
Questions to put to your teams:
When a team finishes work faster, who decides how the saved time is used? Have targets been reset upward without anyone saying so? Do we track errors and rework, or only output? Who is working through lunch and evenings, and is anyone asking them to?
Faster work only saves time if someone decides to keep it.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: UC Berkeley Haas School of Business.
