Capital region absorbed 88.3% of Korea's agentic-AI job growth last year

Compare the share of the gains with the share of the base. The capital region already holds 71.9% of Korea's knowledge service workers and 74.4% of its semiconductor manufacturing workforce, according to the Bank of Korea ^[12]. It then took 80.8% of the three-year employment growth in occupations highly exposed to generative AI ^[5] and 88.3% of last year's growth in agentic-exposed occupations ^[6]. That is 8.9 points ahead of the knowledge-services base on the first measure ^[3] and 16.4 points ahead on the second ^[4]. The note does not include the capital region's share of national employment ^[16]. That would be the cleaner benchmark.

What was left for everywhere else is small. Outside the capital, generative-exposed occupations added 48,000 jobs across the three years, about 16,000 a year ^[1], and agentic-exposed occupations added 12,000 last year ^[2]. The two figures cover different periods, so they do not sum. Inside the Seoul area, the higher an occupation's AI exposure, the faster its employment grew; outside it, the central bank found no clear relationship ^[7].

Physical AI is where the note moves from measurement to forecast. Employment in occupations highly exposed to physical AI has been falling for a long period in both the capital region and outside it ^[8]. The warning is conditional: the Bank of Korea said losses could materialise in regions where manufacturing, accommodation and food services, and transportation carry large employment shares, if the technology advances more quickly ^[9]. On exposure, the top of the table is North Gyeongsang, South Jeolla, North Chungcheong and Ulsan ^[4].

The age numbers complicate the story about Seoul winning. Employment in AI-exposed occupations rose across age groups overall, but among people in their 20s it fell for occupations highly exposed to both generative and agentic AI, inside and outside the capital ^[10]. The Bank of Korea attributes that to AI raising demand for expertise and hands-on experience, which it says could become a burden for young people entering the labour market ^[11]. The 199,000 and the 93,000 accrued to older cohorts.

I'd expect the gap to widen, and the counter-case sits in the same note. The narrowing route is real in principle: digitising human experience and knowledge reduces the need for face-to-face contact and geographic clustering, and pairing it with physical AI could push manufacturing outside the capital toward knowledge-based services ^[14]. It requires adoption the employment series has not registered yet. If the agentic figure drifts back toward 71.9% as adoption broadens past the early users, the widening thesis fails; if next year prints above 88.3%, it holds. The remedy the note reaches for is institutional, and the Bank of Korea recommended that "regional flagship universities be developed as AI hubs to connect local talent with local companies, and that knowledge services and startups linked to manufacturing be fostered" ^[15].