Deep Ideas
← Back to Blog
Blog

Working Less, Working Smarter: How AI Is Rewriting the Global Labor Story

Deepanshu

A few weeks ago I wrote about Jevons paradox and the popular AI narrative around it. The idea that making something more efficient often leads us to consume more of it, not less. That got me wondering about the flip side: what is AI actually doing to work itself? Are we finally seeing efficiency gains translate into fewer hours worked, or are we just going to fill the extra capacity with more work, the way Jevons paradox would predict?

For most of the last 40 years, the story of work followed a predictable curve, fewer hours per person, more people entering the workforce, and slow, steady productivity gains. Then AI showed up and in just the last five years, that curve has started to go in new directions, in ways that feel directly relevant to that Jevons paradox question.

Here's what the numbers say, and what's changed since AI tools went mainstream.

The Four-Decade Trend: Fewer Hours, More Participants

Using OECD, ILO, and World Bank labor data as a baseline, two long-run trends stand out across most developed economies since the 1980s (used AI to compile ofcourse :) ):

Average annual hours worked per employed person have declined almost everywhere.

Country          ~1985     ~2024     Change
Germany          ~1,580    ~1,340    -15%
Japan            ~2,100    ~1,610    -23%
South Korea      ~2,800    ~1,900    -32%
United States    ~1,830    ~1,800     -2%
United Kingdom   ~1,780    ~1,530    -14%
France           ~1,650    ~1,510     -8%

Germany, Japan, and South Korea show the steepest declines, largely driven by stronger labor protections, shorter workweek norms, and automation in manufacturing. The U.S. is the outlier: hours worked have barely moved in 40 years.

More numbers to note, globally, participation rates dipped slightly (from 65% in the 1980s to 62% today, per ILO estimates).

The Last Five Years: A Sharper, Faster Curve

Zoom into 2019-2024 and the trend lines get crazy. Pandemic disruption, remote work, and now AI adoption are compressing decades of change into a handful of years. Two country-level stories make this concrete:

South Korea. Hours worked already fell 32% between 1985 and 2024 which is the steepest decline of any major economy. What's new is the pace of the last five years alone: South Korea's average annual hours have fallen a further 8% since 2019. AI-driven automation in manufacturing and robotics-AI convergence accelerated the decline that had already been running for four decades.

United Kingdom. The UK saw a 14% decline in average hours from 1985 to 2024. Since 2022, four-day workweek programs in the UK (alongside similar programs in Iceland and parts of Japan) have cut hours by a further 10-20% for participating organizations while productivity held steady. It’s interesting to note that that AI and new work arrangements are reducing hours, but it’s more interesting that they're doing so without a corresponding drop in output.

Where Hours and AI Intersect

1. Countries with high AI adoption are seeing the fastest recent drop in average hours worked. South Korea and the UK are both in the top 5 for AI adoption growth and also show some of the sharpest declines in average hours since 2019, outpacing their own 40-year trendline. This is the same acceleration visible in the country-level numbers above.

2. Participation rates are the same, not declining. Despite fears of AI-driven job loss, labor force participation in the top 10 AI-adoption countries has stayed flat or risen slightly since 2022. AI appears to be reshaping how hours are spent more than reducing who is working. This matters directly for the worry that fewer hours automatically means fewer jobs so far, the data doesn't support that claim.

3. Productivity per hour is rising faster where AI adoption is highest. Early data (OECD, 2024) suggests labor productivity growth in high AI adoption countries (US, South Korea, Singapore) is running 1.5-2x higher than in low-adoption countries over the past two years, a gap that barely existed pre-2020.

What This Means

The 40-year curve was not sharp and the change was slow. The last five years that are supercharged by AI, look more like compression than continuation. Countries are fitting decades' worth of change into a handful of cycles.

This brings us back to the Jevons paradox question I started with. If AI is making work more efficient, the paradox would predict we simply do more work with the freed-up capacity, more output, more hours, more consumption of "work" as a resource. But the South Korea and UK data suggest something different is happening, at least for now, hours are falling and productivity is improving. A reason for cautious optimism rather than a full escape from the paradox is that five years isn't long enough to call it a permanent pattern.

The second open question is whether fewer hours worked will eventually translate into fewer people working (a shrinking participation rate and fewer jobs overall). The data through 2024 doesn't show that yet, participation in high-AI-adoption countries has been flat to slightly up since 2022, even as hours have fallen. But that's a short window, and it's worth watching closely rather than assuming it holds.

So the open question is whether this becomes a permanent decoupling of hours from output (fewer hours, same or more value created, without a hit to jobs) or a temporary spike before a new equilibrium settles in, one where hours and headcount both eventually fall together.

Sources:

  • OECD Average Annual Hours Worked: https://stats.oecd.org/Index.aspx?DataSetCode=ANHRS
  • ILO Labour Force Participation Data (ILOSTAT): https://ilostat.ilo.org/data/
  • World Bank Labor Force Participation Rate: https://data.worldbank.org/indicator/SL.TLF.CACT.ZS
  • Stanford HAI AI Index Report 2024: https://aiindex.stanford.edu/report/
  • McKinsey Global Survey on AI: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai