2026-08-29 Weekly Reading: With Stronger Technology, Why Are People Not Working Less?

2026-08-29 Weekly Reads: Tech Gets Stronger, So Why Aren't People Working Less?

Why This One

This week's WeChat forwards kept circling back to "stages of society": Is gig work a benefit or a trend? Will the one-person company replace traditional organizations? Will AI let young people skip the accumulation phase? And can individuals change their fate by changing their behavior? These discussions share a premise that's easy to overlook: after productivity rises and technology upgrades, people should naturally be freed from labor.

The research by Amory Gethin and Emmanuel Saez is worth reading precisely because it doesn't accept that premise upfront. They pieced together labor force surveys from 160 countries into a comparable micro-database, with cross-sections covering 97% of the global population, and constructed time series spanning over 20 years for 86 countries. The result isn't "advanced countries necessarily work less," but a more uncomfortable and more explanatory finding: total market labor time for prime-age adults has been remarkably stable over the long run. Development mainly changes who does the labor, and which groups institutions allow to exit the labor market.

This connects to what's genuinely valuable in "analyzing stages of society": you can't just slap a version label on an era. You have to find how the version lands in observable institutions—schools, pensions, formal labor contracts, overtime rules, and household division of labor.

Core Argument of the Paper

The research first standardizes the metrics. The authors count all work that goes into GDP, including wage employment, self-employment, and unpaid agricultural work producing goods, but they don't count household services like cooking, cleaning, childcare, and elder care in the main database. In the latest available cross-section, 59.3% of the global population aged 15+ is employed; employed people average 41.5 hours per week, which spreads to 24.5 hours across all adults. Men contribute 65% of market labor time, women 35%, and the gap comes mainly from employment rates, not just weekly hours of those employed.

Breaking down by age, "development makes people work less" only clearly holds at the two ends. Labor time for 15–19-year-olds falls with development, mainly because more people enter school; labor time for those 60+ falls, mainly because pensions make exiting the labor market possible. Income level itself isn't the full explanation—education and retirement systems are the mechanisms that directly carry the change.

Prime-age adults aged 20–59 show a different picture. Whether comparing countries at different development levels cross-sectionally, or looking at multi-decade time series across countries, per-capita market labor time for prime-age adults is remarkably stable. Prime-age Americans average about 30 hours of market labor per week—virtually unchanged from 1900 to today. Long-run series from other regions show similar stability. Technological progress hasn't mechanically squeezed down this chunk of total labor.

Underneath the stable total sits a massive gender reorganization. During development, weekly hours for employed prime-age men fall, while women's labor force participation rises—two forces that almost exactly offset in many countries. Men's change happens mainly in "how long each employed person works"; women's change happens mainly in "whether to enter market work at all." So total hours look unchanged, but who does the labor has shifted.

The main database excludes housework—does that make the gender reorganization a statistical illusion? The authors test this with time-use surveys from 84 countries: with development, women's household labor falls and men's rises, exactly opposite to the market labor direction. Including household production, men's and women's respective total labor time remains broadly stable with development. The more accurate statement isn't that women suddenly started working more, but that labor is being redistributed within both the market and the household.

The paper ends with a discussion of taxes, welfare, and labor regulation. In cross-country data, countries with higher labor tax rates have shorter working hours, but this relationship weakens significantly once social spending is added; after adding an index of formal-sector coverage and working-time regulation, the correlation between tax rates and hours approaches zero. The authors argue that high taxes alone aren't a sufficient explanation—they tend to appear alongside social welfare, maximum hours, overtime, and vacation rules. Institutions convert productivity into a certain kind of life; productivity doesn't automatically make the choice for people.

Where to Push Back

First, the research is strongest on description and comparison, not on pinning down causal chains completely. Education, pensions, social spending, labor regulation, formal employment, and cultural norms all change together, and cross-country regressions can't prove any single one causes the hours change. The "effective regulation index" the paper uses is itself a noisy proxy, as the authors admit. Seeing the tax coefficient disappear after adding regulation only shows the simple "high taxes cause less work" narrative doesn't hold—it doesn't directly imply any specific regulation will reduce hours by a certain amount.

Second, harmonizing 160 countries' surveys inevitably sacrifices detail. Countries differ on actual vs. usual hours, all work vs. main work, weekly vs. longer recall periods, and self-response vs. proxy response. The authors run robustness checks on seasonality, recall periods, and multiple jobs, and the overall conclusions don't change materially—but precise rankings for individual countries shouldn't be over-read. China's main data still uses recall of hours over the past year, though it's close to the 30–31 weekly hours from two surveys using weekly recall; the measurement methods still aren't fully homogeneous.

Third, defining "work" by national accounts boundaries misses a huge amount of care and housework not counted in GDP. The 84-country time-use surveys support the stable total labor time conclusion, but they can't provide the same 160-country coverage and historical depth as the main database. If you care about exhaustion, free time, or family pressure rather than market production, the global average of 24.5 hours severely underestimates the real burden.

Fourth, global averages hide distribution. Stable prime-age total hours don't mean the same people have stable jobs, nor that job quality, income, control, and security are stable. Platform labor, multiple part-time jobs, unemployment, and overwork can all coexist in a polarized way under the same mean. The paper answers "has the total automatically declined," not "is everyone better off."

What This Means for Recent Concerns

This research adds a necessary brake to "stages of society": stage changes aren't natural laws, and seeing AI get stronger doesn't let you infer "companies disappear, everyone self-employed, hours fall." Productivity only expands the option space; the final outcome passes through contracts, education, welfare, organizational power, and household division of labor. Without these transmission mechanisms, the so-called next stage is likely just the same amount of labor with a different bearer.

For individuals, it also corrects the over-individualized version of "change your fate by changing your behavior." Behavior matters, of course, but whether someone can exit low-value labor, extend their study period, take entrepreneurial risk, or care for family is genuinely constrained by pensions, healthcare, family responsibilities, and labor contracts. Explaining all outcomes as willpower misses institutional costs; explaining all outcomes as the tide of the times erases individual choice. A more practical approach is to separate the two layers: what's your own work portfolio and skill accumulation that you can change, and what requires cash buffers, insurance, contracts, and family negotiation to change.

"One-person company" also doesn't mean one person actually works less. AI can replace certain tasks, but it may push client acquisition, delivery, after-sales, compliance, and risk all back onto the same person. Evaluating this model shouldn't just look at headcount or tool speed—you should at least track total weekly hours, uninterruptible responsibilities, income volatility, rework, and manual firefighting. If output rises but time doesn't fall, that's labor boundaries being repackaged, not freedom gained.

The research also imposes a very plain constraint on future predictions: don't write technical capability directly as social outcomes. Between "what AI can do" and "how long people ultimately work" sits how organizations distribute gains, whether workers can refuse extra tasks, and whether public institutions provide exit options. When judging version changes, track these intermediate variables, not just the model capability curve.

How to Read It

Start with the authors' four-page overview in the IMF's Finance & Development, and grasp the three-way splits: all adults vs. employed people, prime-age vs. youth and elderly, total vs. gender division. Don't memorize country rankings first.

Then open the full paper, focusing on the database definitions, the sample expansion in Figure 8, the gender reorganization, and the tax-and-regulation regressions in Section 6. For every conclusion, ask: is this a cross-sectional correlation, a time-series fact, or causal identification? In particular, don't rewrite "the tax correlation disappears after adding regulation" into a definite policy effect.

Finally, apply it to your own work log: for four consecutive weeks, record actual hours of market work, unpaid housework, study, and manual firefighting. Then look at whether AI tools are actually reducing total time, increasing output, or shifting tasks from the team onto you. The best use of macro research isn't to draw conclusions for individuals—it's to teach people to choose metrics that don't fool themselves.

Authors' overview: Gethin and Saez: How Much Does the World Work?

Full paper and reproducible data portal: Gethin and Saez: Global Working Hours