One Last Prompt Download
← Blog

AI was supposed to save you time. Where did the evening go?

Zsolt Csizmadia, founder · · 3 min read

The promise was simple: the same work in less time. The early evidence describes something else.

I use AI tools every day, and this is not an article against them. It is about a different question: where the saved time went. For me it did not go to the evening.

One company, eight months

Two researchers from UC Berkeley spent eight months at a technology company of about 200 people where AI tools were available to everyone.1, 2 They were on site and did more than 40 interviews. They describe three changes. People took on a wider range of tasks. The natural pauses of the day wore away: prompts were sent during lunch, before meetings and in the evening. And several AI threads ran in parallel. The employees said they felt busier, more stretched and less able to disconnect.

It is research in progress at one company, with no measured hours. It is also the closest description I have read of my own days.

The time-diary data

Economists looked at the same question with numbers.3 Using time diaries of American workers from 2004 to 2023, they found that occupations more exposed to AI show longer workdays and less leisure. Their reading is that AI adds to what a person can do instead of replacing the person, and that the effect is stronger where workers have less bargaining power.

Two cautions. It is a working paper that has not been through peer review. And most of those years are from before chat assistants, so “exposed to AI” means something broader than the tools I use.

Developers

A survey of 442 developers linked the adoption of generative AI to more burnout, through higher demands of the job.4 Good resources at work and a positive view of the tools weakened the link. It is a preprint and a single survey, so it shows a connection and not a cause.

Researchers say themselves that the evidence is thin. A position paper from 2026 argues that these tools can add mental load and checking work and raise the expected pace, and that studies should measure the well-being of developers, not only their speed.5

The ground it lands on

The workday had already lost its edges before any of this. Microsoft’s report on activity in its own office software found that meetings after 8 pm were up 16% in a year, that the average employee sends or receives more than 50 messages outside core hours, and that by 10 pm 29% of active workers are back in their inbox.6 That is email and meetings, not AI, and it is one company’s own data. It is the day that the new habit arrives into.

The saved time did not go to the evening. It went to the next task.

Why a faster tool makes a longer day

Here is my own explanation, and it is only that. A tool that makes starting cheap removes the reason to wait. Every idea can be tried now, so every idea becomes an open thread, and open threads are what follows you into the evening. The work per task went down. The number of tasks went up by more.

What nobody has measured

I have found no study of founders, freelancers or people who build alone. The evidence is one company, one working paper, one survey and many personal reports, mine among them. It may look different in two years.

That is why I did not wait for the research to settle. If a tool takes away the natural end of the day, the end has to be set by hand. I set mine as a closing time, and built an app that keeps it.

“One more prompt” becomes one last prompt.

One Last Prompt is an iPhone app, and it is out now on the App Store.

Download on the App Store

Zsolt Csizmadia builds One Last Prompt.

Sources

  1. Aruna Ranganathan and Xingqi Maggie Ye, “AI Doesn’t Reduce Work—It Intensifies It”, Harvard Business Review, 9 February 2026. hbr.org
  2. Laura Counts, “AI promised to free up workers’ time. UC Berkeley Haas researchers found the opposite”, Berkeley Haas Newsroom, 18 February 2026. newsroom.haas.berkeley.edu
  3. W. Jiang, J. Park, R. Xiao and S. Zhang, “AI and the Extended Workday: Productivity, Contracting Efficiency, and Distribution of Rents”, NBER Working Paper 33536, February 2025. nber.org
  4. Z. Feng, S. Afroz and A. Sarma, “From Gains to Strains: Modeling Developer Burnout with GenAI Adoption”, arXiv:2510.07435, preprint, 2025. arxiv.org
  5. M. Guizani, M. Subasinghage, S. A. Licorish and S. Ouhbi, “At What Cost? Software Developers’ Well-Being in the Age of GenAI”, arXiv:2605.22349, position paper, 2026. arxiv.org
  6. Microsoft WorkLab, “Breaking down the infinite workday”, Work Trend Index Special Report, 17 June 2025. microsoft.com