Does AI make you work more? What the data says about AI and longer hours

Does AI make you work more? Orestis Ioannou, who leads engineering at Bold.org, says it does for him. His essay "AI Promised Less Work. I Have Never Worked More." was published on October 27, 2026, and reached 630 points and 626 comments on Hacker News that day under the title "AI promised less work. I've never worked more." The research on AI and jobs gives a split answer. The largest studies find no measurable change in paid or recorded hours, but surveys, a time-use study and close fieldwork point to faster, fuller and sometimes longer workdays. Each kind of study measures something different, and none settles the question alone. The same gap between headline and measurement runs through Anthropic's claim that agent hours outnumber its employees' six to one.

This analysis is based on the essay, NBER Working Paper 33777 and its abstract, the St. Louis Fed's Real-Time Population Survey data on FRED, a working paper on the 2026 American Economic Association conference site, Berkeley Haas and Harvard Business Review reports on a field study, Upwork's July 2024 survey release, Microsoft's June 2025 Work Trend Index report, METR's February 2026 update and Bureau of Labor Statistics hours data, checked October 28 and 29, 2026.

The short version

• The essay is one person's account, backed by his GitHub activity, which he calls "a rough signal of activity, not a measure of value."
• Paid hours haven't moved much. A study of Danish workers rules out effects on recorded hours larger than 2%, and US average weekly hours were 34.4 in September 2026, against 34.5 when ChatGPT launched.
• Self-reported time savings are small: about 2.1% of all US work hours in the third quarter of 2026, according to the St. Louis Fed's tracker.
• The case for "more work" rests on other kinds of evidence: a correlation in time-use diaries, an eight-month study of one company and a survey commissioned by a freelance marketplace.
• The likeliest reading is that AI changes what fills the workday before it changes how long the paid workday is.

What does the "AI promised less work" essay argue?

Ioannou opens with the promise: "The pitch was that AI would give us our time back." His account is that agents made implementation cheap, so he now runs several at once and takes on far more. "Because implementation is now cheap, there is no natural friction stopping me from pursuing all of them at once," he writes, and "working with many agents is tiring in a new way." He describes 2026 as a series of bursts, each set off by a new model or workflow, followed by a pullback when reviewing and context switching catch up.

His evidence is his own GitHub contribution calendar. He writes that "August 2026 is the busiest month in the whole timeline," with 427 contributions against a previous high of 319 in April 2020, and that his median weekday rose from 3 or 4 contributions in 2022 to 2025 to 7 in 2026. Through October 27 he had 18 weekdays with no contributions, against 40 to 46 in each full year from 2022 to 2025.

He is careful about what that shows. Contributions mix commits, pull requests, reviews and issues and include private work. They count activity, not hours. His role also changed over the period: he writes that he was a principal engineer until August 2025 and has been vice president of technology since. His conclusion is a personal one: "being able to do more keeps turning into doing more."

What do the largest studies say about AI and working hours?

The biggest study to date finds no effect on recorded hours. Economists Anders Humlum and Emilie Vestergaard link large surveys on chatbot adoption to Denmark's administrative labor records. Their NBER paper, first issued in May 2025 and revised in March 2026 under the new title "Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI," says "most employers in exposed occupations have adopted chatbot initiatives." Yet the authors "estimate precise null effects on earnings and recorded hours at both the worker and workplace levels," "ruling out effects larger than 2% two years after the launch of ChatGPT."

What changes instead, they write, is the content of jobs: "employers absorb AI through task reorganization," including "new tasks in content generation, AI oversight, and AI integration." Their summary line fits the essay surprisingly well: "Technological change reshapes work well before it surfaces in earnings or hours."

The limits matter for this question. Recorded hours in administrative data track contracted or registered time, not evenings spent reviewing an agent's output. The study covers Denmark, a country with strong labor protections, and its window ends about two years after ChatGPT's launch.

US payroll data tell a similar story at the national level. The BLS's average weekly hours for all private employees were 34.5 in November 2022, the month ChatGPT launched, and 34.4 in September 2026 (preliminary), close to their 2019 level. That series counts paid hours per job. It can't show unpaid overtime or after-hours messages, and it can't separate AI from everything else that moves hours.

Does AI actually save time at work?

A little, by workers' own accounts. Economists Alexander Bick, Adam Blandin and David Deming run a nationally representative online survey of US adults aged 18 to 64, published by the St. Louis Fed. Its time-savings series, the share of all work hours saved by generative AI, was 2.08% in the third quarter of 2026, up from 1.6% in the third quarter of 2024. Non-users count as zero, so the figure is much higher for people who use AI: in the November 2024 survey, users reported saving 5.4% of their work hours, about 2.2 hours of a 40-hour week.

Those savings are self-reported answers to a hypothetical question about how many more hours a task would have taken without AI. The authors note that workers may spend saved time as on-the-job leisure, which "would increase welfare but not productivity." The survey doesn't ask whether saved time turns into extra tasks, which is the essay's claim.

Is there evidence that AI makes the workday longer?

Some, from three different kinds of study, each with clear limits.

Time-use diaries. A working paper by Wei Jiang, Junyoung Park, Rachel J. Xiao and Shen Zhang, listed on the 2026 American Economic Association conference site (current draft December 17, 2025), uses American Time Use Survey diaries from 2004 to 2023. It finds that AI exposure "is associated with longer work hours and less leisure." Around ChatGPT's release, "an interquartile increase in generative AI exposure corresponds to an additional 3.75 hours" of work a week, with about 3.85 fewer leisure hours. The paper measures exposure by occupation, not actual AI use, it is unpublished, and its estimates are associations rather than proof of cause.

Fieldwork at one company. Berkeley Haas researchers Xingqi Maggie Ye and Aruna Ranganathan spent eight months at a US technology company of about 200 employees, observing work and conducting more than 40 interviews. Writing in Harvard Business Review on February 9, 2026, they reported that "employees worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day." They describe AI use spilling into breaks and evenings and people juggling several AI-assisted tasks, including multiple agents, at once. Berkeley Haas calls the research in progress, and one company can't stand in for the economy.

A vendor-commissioned survey. Upwork's research institute, with Workplace Intelligence, surveyed 2,500 people in the US, UK, Australia and Canada from April 16 to May 5, 2024: 1,250 executives, 625 full-time employees and 625 freelancers. Upwork said "77% of employees using AI say these tools have added to their workload," while 96% of executives expected AI to raise productivity. The employee sample is small, the data are two years old, and Upwork, a freelance marketplace, framed the findings as a case that "introducing new technologies into outdated work models and systems is failing."

Last verified: October 29, 2026

Why would saved time turn into more work?

The sources offer several mechanisms, none proven at scale:

• Lower friction. Ioannou's point: when trying an idea costs little, people try more of them. Ye says "a vicious cycle can form" of higher output, rising expectations and further expansion.
• New tasks. The Danish study finds employers adding work in AI oversight and integration, which could absorb time saved elsewhere.
• Fragmented attention. Microsoft's June 2025 Work Trend Index report, based on Microsoft 365 signals and a survey of 31,000 workers, found 40% of employees checking email before 6 a.m., meetings after 8 p.m. up 16% year over year, and interruptions every two minutes. Microsoft didn't blame AI for this "infinite workday"; it pitched AI agents as the fix.
• Hard-to-measure time. METR, which ran a widely cited 2025 trial finding experienced developers took 19% longer with AI tools, said on February 24, 2026, that time tracking becomes unreliable when developers run multiple AI agents at once, one reason it is redesigning the study. Our guide to coding-agent evaluations covers that trial.

So does AI make you work more?

What the evidence shows as of October 2026: AI hasn't measurably lengthened paid or recorded working hours in the national data studied so far, from Denmark and the US, but there is credible, if limited, evidence that it makes work more intense and lets it spill into more of the day for some people, especially those who use it heavily. Self-reported time savings are real but small, and nothing in the data shows where the saved time goes.

That fits a pattern seen elsewhere on the jobs beat. The IMF's 40% exposure figure measures overlap with AI, not job loss, and the data behind one count of 200,000 lost creative jobs can't say how much is AI. Hours data have the same blind spot: they count time paid, not effort spent.

What's confirmed, reported and uncertain

• Confirmed in primary data: the Danish null result on recorded hours (NBER abstract), the BLS hours figures and the St. Louis Fed time-savings series.
• Reported by researchers, with stated limits: the time-use association (unpublished working paper), the Berkeley Haas fieldwork (one company, in progress) and METR's account of its measurement problems.
• Vendor-commissioned or vendor-run: Upwork's 77% figure and Microsoft's infinite-workday data.
• One person's account: Ioannou's essay and his GitHub numbers.
• Uncertain: whether heavy AI users work longer unpaid hours on average, and whether any intensification lasts once tools and habits settle.