The case files.

An analysis of France's 1998–2002 shift from 39 to 35 hours with pay preserved estimated that payroll tax cuts offset job losses from the hours cut, leaving employment largely unchanged; total factor productivity rose about 2% and output fell about 1%. This working paper studies firms around a size-based legal threshold, and its results concern the combined reforms, not a 32-hour week or AI.

Two years after a German pilot, 70% of 40 responding organizations still used reduced hours (including 22% of all respondents who had adapted their model), while 30% had stopped. Continuing organizations reported work-life benefits, while those that stopped reported transition costs; these are self-reported perceptions, not evidence of causal effects or audited productivity gains.

Iceland's 2015–2019 public-sector trials cut many participants' weeks from 40 to 35–36 hours without pay cuts, and the report says productivity and services were maintained or improved in most workplaces. The trials covered over 2,500 workers in more than 100 workplaces, had no control group, and did not test a 32-hour week; some later reductions required extra staffing costs.

A six-month shorter-week study found improved self-reported well-being among 2,896 employees in 141 organizations, a pattern absent in 12 control organizations. Nonrandomized, income-preserving trials in six countries; schedules and hours varied.

Published estimates put US annual working hours per worker at about 3,100 in 1870 and 1,790 in 2023, with large declines also shown for Germany and the UK. Historical figures cover full-time production workers, while post-1950 averages include the whole economy and part-time work; these annual estimates are not a uniform measure of weekly hours.

Two Scottish public bodies with 259 staff trialed a 32-hour full-time week without pay cuts for one year, broadly maintained internal service metrics, and reported improved staff wellbeing; both extended the pilot. This was a before-and-after evaluation without a control group; wellbeing was self-reported, and the results do not show that every workplace could achieve the same outcome.

A six-month pilot in 11 Swedish and Norwegian organizations reduced working time with full pay and was associated with less stress, better sleep and higher self-reported productivity; 10 organizations continued. Schedules varied, including shorter days, and the before-and-after study had no control group, so it does not establish causal effects or a uniform 32-hour week.

In a randomized online experiment with 1,174 Argentine adults, AI assistance improved scores on a short business problem-solving task and narrowed the gap between higher- and lower-education participants. Participants were aged 25–45, and the study measured performance on one task outside a workplace, not wages, employer productivity or long-term learning.

Across three randomized workplace experiments involving 4,867 software developers, researchers estimated that GitHub Copilot use increased weekly completed tasks by about 26%. Results varied across the three large employers, and the estimate concerns use of the tool rather than access alone; the study did not measure code quality, pay or working hours.

In a six-month experiment involving 7,137 knowledge workers at 66 large firms, Microsoft 365 Copilot access cut Outlook email time by 1.4 hours per week and after-hours app use by 15 minutes per week. The randomized study detected no change in the number of email threads, meetings or documents; it measured activity in Microsoft applications, not output quality or overall productivity.

In a randomized trial involving 16 experienced open-source developers and 246 real issues, tasks with AI allowed took 19% longer to complete than tasks with AI disallowed. The study used early-2025 tools in large, mature repositories and measured completion time; it does not show how all developers, other tasks or later AI tools perform.

AI assistance increased issues resolved per hour by 15% on average in a study of 5,172 customer-support agents at one firm and its contractors. Mostly Philippines-based agents serving US customers; effects varied by experience.

In a survey of 1,500 frontline shift workers in the US, UK and Australia, 15% said they were always consulted before new AI tools were introduced, while 42% said they were rarely or never consulted. The vendor-commissioned online survey covered hospitality, foodservice, retail and healthcare; responses were self-reported and may not represent all workers.

US net productivity grew 93.7%; typical workers' hourly compensation grew 32.7% from 1979 Q4 to 2026 Q2. These are inflation-adjusted US series: compensation includes wages and benefits for private-sector production and nonsupervisory workers, while productivity covers the total economy.

Across 52 high-income countries, average labour productivity grew 29% from 1999 to 2024, while average real wages grew 15%. These population-weighted figures compare GDP per worker with inflation-adjusted average wages, not typical workers' pay; 2024 uses first-half data, and most of the gap opened before 2007.