The 94 per cent and the number beside it
The Bank of Korea's employment research team published a report on 18 August with a figure built to travel: between June 2022 and June 2026, jobs held by people aged 15 to 29 fell by 285,000, and 268,000 of that decline landed in sectors with high exposure to AI. The sector detail is sharp. Youth employment in information services fell 31.4 per cent over the four years, publishing 27.4 per cent, computer programming and systems integration 16.6 per cent, professional services 11.6 per cent. Read alone, the concentration invites one conclusion.[1]
The report puts another number next to it. Over the same four years and in the same high-exposure sectors, jobs held by people in their fifties rose by 230,000, and 173,000 of that increase — 75.2 per cent — came in exactly the sectors where the young were disappearing. The thinning in publishing and information services did not run through those sectors' total appetite for labour; the same sectors were taking on workers in their fifties. Two movements ran in one place, in one window, in opposite directions.[1]
Which rung the automation reached
The bank offers a mechanism rather than a slogan. Young workers do a larger share of work driven by a manual; experienced workers do a larger share of work that rests on the organisation's context, so the first wave of adoption reached them less. The split inside the data supports that reading: the youth decline is marked where firms hand a task people used to do over to AI, and unclear where AI drafts and checks alongside a person. Education tracks the same edge. After ChatGPT arrived in November 2022, average unemployment among four-year graduates ran at 7.0 per cent against 5.4 per cent for junior-college graduates and below, a gap of 1.6 percentage points; between January 2019 and October 2022 the same two figures were 8.2 per cent and 8.0 per cent. Among workers with five years of experience or less, measured exposure rises with the diploma: 0.23 at high school, 0.25 at junior college, 0.27 at four-year level and above.[1]
The flow numbers say the same thing from the other side. Monthly average youth departures from high-exposure sectors rose from 3,700 in the 2016 to 2019 period to 4,900 between July 2022 and June 2026, an increase of 32 per cent, while monthly entries fell from 32,600 to 29,100, a drop of 11 per cent. Both doors moved. Fewer people got in, and more of those already inside left. A career ladder needs a bottom rung to hold weight, and this is what it looks like when the rung takes less.[1]
What does a vendor's ledger show, and what can it not show?
On 17 August Salesforce put numbers on the shape of that work, from its own platform and therefore as an interested account. Across 400 businesses and five quarters of Agentforce data, the average organisation went from 5 agents in production in February 2025 to 13 in April 2026, and the time to put a new agent to work fell 53 per cent, from 4 days to 1.9. Unique actions per agent rose from 2 to 4. In early 2024 no customer service session was closed by an autonomous agent; the company now counts 7 in every 10. On its own help site in August 2026 the volume handled by agents passes 5 million conversations, against 2.4 million handled by people. Answering routine service questions is not a marginal job category. It is one of the places a first job used to be.[2]
So the honest reading is narrower than the headline and more uncomfortable than a denial. A vendor's platform in one market cannot explain an employment series in another, and the bank names its own competitors for the same variance: post-pandemic sectoral adjustment, hiring that favours experience, the spread of remote work. What the two sources share is a direction — the work being handed over is the manual-driven kind, and that is the kind a career used to start with. When I argued on 15 August that thinning entry-level work drains the pool of people who can later catch a model's mistakes, the objection was that no population-level evidence existed. This is population-level evidence, and it is not yet a verdict. The measurable test is the entry flow itself: if monthly youth entry into high-exposure sectors is still below 32,600 when the bank next updates this series, the ladder has not been rebuilt, whatever the retraining budget says. The bank's own recommendation concedes the point by asking for a new ladder rather than the old rung back.[1], [2], [3]