AINEWS
Korean workers log 5–7 hours a week with AI, roughly twice the US figure, yet save less time. The data points to AI crossing from tool into environment.
Korean workers spend five to seven hours a week inside AI tools, roughly twice what American workers spend. They save less time for it.
The 51.8% of Korean workers using generative AI on the job, against an estimated 26.5% in the US, comes from a Bank of Korea household survey of 5,512 respondents published in August 2025 as BOK Issue Note 2025-22. The American comparison traces to a late-2024 NBER working paper, cited in trade press. Heavy users, defined as anyone spending more than an hour a day with AI, are 78.6% of Korean workers and 31.8% of American ones. Korea adopted faster and stayed longer.
If AI were a productivity tool, the country using it most would save the most time. Korea uses more AI and saves less time. The BoK Issue Note estimates AI reduces Korean working hours by 3.8%, about 1.5 hours off a 40-hour week. The American figure, by the same comparison, is 5.4%. The inversion holds at every adoption tier the survey measures.
A Bank of Korea-funded preprint by Suh, Oh, and Kim of the BoK Employment Research Team explains the mechanism. The correlation between time saved through AI and output produced is near zero. Efficiency gains get captured as on-the-job leisure: browsing, smaller tasks, slower throughput, rest built into the working day. The authors call this time reallocation rather than time savings, and the distinction matters. Savings implies the time goes somewhere productive. Reallocation means the time stays inside the worker, mostly invisible to the firm.
A separate Now&Survey of 1,000 Korean office workers, surfaced in the inlevel9 letter that collected this body of data, gives consumer-side confirmation. 54.1% of heavy users say their working hours have not actually decreased despite regular AI use. 61.2% say their organization encourages AI use. Encouragement, hours, no reduction.
Trajectory data shows the pattern is moving, not staying in Korea. Seoul Economic Daily and Chosun tracking shows Korean GenAI usage rising from 33% to 44% in roughly three months through early 2026, and Korea leading the world in 2026 AI-adoption growth at +6.4 points year over year. The country has crossed from early majority to something denser, where AI is no longer a tool a worker opens but a layer they sit inside for most of the working day. Once the instrument is already in the worker's hand, the "use it carefully" advice that works for a tool stops working too.
Most American and European workers are on the same curve, just earlier on it. Their hours-per-week figure, between half an hour and two hours, is roughly a third of Korea's. Within two years, the survey authors and the diffusion math both suggest, that ratio closes. The questions Korea is confronting now become familiar ones: where does the saved time go, who captures it, and what is the difference between a workforce that uses AI a lot and a workforce that uses it well.
Three moves are worth making before the curve catches up. Audit your own use: does a typical AI session end with output, or with time absorbed? If time absorbed is rising while output is flat, you are already on the Korean pattern at lower volume. Ask whether your employer measures hours worked, output, or both. The Korean inversion is invisible in any firm that only tracks attendance. Treat the "use it carefully" advice as a tool-era instruction and push for environment-era design: scheduled deep-work blocks, output reviews, and explicit policy on what AI use is for. Korea's data does not say the technology failed. It says the frame did.