AI and work in Japan

The Anglophone debate about AI and work is an argument about whether machines will take jobs. In Japan that argument does not apply, and watching a country where it does not apply is the fastest way to see how much of the Western discussion is local rather than universal.

Japan has roughly 29 per cent of its population over 65, unemployment near 2.5 per cent, a fertility rate around 0.75, a working-age population that peaked in 2019, and a projected shortfall of about 11 million workers by 2040. There is no plausible scenario in which AI creates a Japanese unemployment problem. The problem is the opposite one.

And here is the finding that should unsettle everyone: Japan has the strongest imaginable economic case for adopting AI, and adoption is running at roughly 18 per cent.

The paradox

If AI adoption were driven primarily by economic necessity, Japan would lead the world. It does not. OECD analysis and Japanese survey work point to a consistent set of constraints, and almost none of them are about the technology.

Firms report a lack of information about the benefits, insufficient examples from comparable companies, difficulty with the data AI training requires, cost concerns, and a shortage of products that are straightforward to adopt. The most pressing reported problem is not a shortage of AI specialists. It is a shortage of employees who can combine workplace experience with basic AI knowledge, which is a description of a capability gap rather than a technology gap.

Structural analysis points further upstream: stagnant competition between firms and weaknesses in higher education, reinforcing each other to suppress investment and limit talent development.

Japanese survey evidence in the SuperSkills evidence base is consistent. Only 12.9 per cent of respondents reported any firm AI use and 8.4 per cent used it themselves, though among users, reports of improved job quality and wellbeing outweighed reports of harm.

What the Anglophone debate is not seeing

1 · Adoption is a capability problem, not a technology problem. The country with the most acute need and substantial state investment, including several billion dollars directed at robotics and AI, is constrained by not having enough people who understand both the work and the tools. That is the argument this research makes, arriving from a completely different direction and without anyone framing it as a warning about capability.

2 · Labour scarcity changes what automation means. Where there are no spare workers, automation is the only way the work happens at all. The evidence base already contains a striking case: Japanese nursing homes adopting robots raised employment and improved retention, most strongly for non-regular staff, reallocating worker effort towards direct care. In a tight labour market, machines can make jobs more attractive rather than fewer.

3 · The displacement question can be irrelevant while the capability question is urgent. Japan can skip the entire jobs argument and still face every question this research asks: who supervises the system, who retains the skill, what happens to the training of the next cohort. Which suggests the capability question is the more fundamental one, and the displacement question is the local expression of a labour market with slack.

4 · Slow adoption is not the safe option it appears to be. A workforce that shrinks by millions while capability to deploy AI stays scarce has an economic problem no amount of caution solves. In Japan, moving slowly is a risk rather than a hedge, which is the reverse of how caution is usually framed in Western commentary.

Why the adoption figures disagree

Adoption figures vary considerably by survey, definition and date, and the 18 per cent and 12.9 per cent figures come from different instruments measuring slightly different things. Treat them as indicating a low level rather than as precise.

The nursing home finding is one working paper in one sector, and robots in physical care work are not language models in knowledge work. Generalising from it would be exactly the error this research criticises elsewhere.

And a limitation to put plainly: this page is assembled from English-language sources, including OECD and IMF analysis of Japan. It is a better view than the Anglophone debate currently has. It is not the view a Japanese-language reading of the primary material would give. That gap is real and is being worked on rather than hidden.

Japan as the control condition

Japan is the control condition the AI-and-work debate never had. Remove the fear of unemployment entirely, add overwhelming economic pressure to automate, and adoption still stalls on human capability.

That is difficult to reconcile with a story in which AI sweeps through economies because it is capable and cheap. It fits much better with a story in which diffusion is governed by organisations, skills and institutions, which is the argument Narayanan and Kapoor make in AI as Normal Technology, and which Japan is currently demonstrating at national scale.

For anyone building a workforce strategy, the practical implication is uncomfortable: the constraint is unlikely to be the tools or the budget. It will be the number of people who understand the work well enough to direct a machine through it, and that number is not increased by procurement. See AI workforce strategy.

Related SuperSkills research

On the capability constraint, capability debt and what AI literacy means for leaders. On the jobs question elsewhere, will AI replace my job and entry-level jobs. On how the discourse formed, the best writing on AI.

Key sources

• OECD. AI use in the Japanese workplace, in Artificial Intelligence and the Labour Market in Japan.
• International Monetary Fund (2025). The Impact of Aging and AI on Japan's Labor Market. IMF Working Paper 2025/184.
• Japan's AI transition faces stumbling blocks. East Asia Forum, June 2026.
• JILPT (2025). Survey on AI use in Japanese workplaces.
• Lee, Y. et al. (2024). Robots and labour in Japanese nursing homes.

About this research

Rahim Hirji is the author of SuperSkills (Kogan Page, 2026) and founder of The SuperSkills Intelligence Company. This page is assembled from English-language sources and states that limitation above. Adoption figures come from different instruments and are indicative rather than precise. Reviewed quarterly.