Two years into the loudest jobs-panic of the AI era, one of the quietest and most useful public projects has started publishing verdicts. A community evidence wiki tracking every major claim about AI and jobs against the underlying US Census household survey, updated on 26 October with data through September 2026, its current answer is the least dramatic sentence in the whole debate: not yet, across the economy — but check the young workers. It reached Hacker News’ front page midweek.
🔍 THE BOTTOM LINE
The broad AI-jobs catastrophe remains, in the data, absent: the 135 occupations most exposed to AI — a quarter of all US jobs — grew 2.8 percent over the past year against 3.7 percent for the rest, a gap smaller than the survey can reliably measure. But the exception is exact and reproducible: 22-to-25-year-olds in the most exposed quarter of occupations saw employment grow just 0.1 percent while young workers elsewhere grew 9.6 percent — and the gap widened to its largest on record (17.9 points, beyond the noise) in the July-September quarter. The AI labour market effect, if it is real, is a door that narrows for entrants, not a bulldozer — and that is a very different career-planning problem.
The economy-wide nil result, and why it took so long to get
The wiki’s core comparison comes from Census CPS microdata mapped against the OpenAI-derived occupation exposure scores in the GPTs-are-GPTs dataset. Its key structural finding: before ChatGPT, the exposed occupations grew faster than the rest; since ChatGPT, the exposed group grew slightly slower — a swing, but at 0.9 points, inside the survey’s noise. Its second check using Anthropic’s usage data found no tilt at all: automation-leaning and augmentation-leaning exposed occupations both changed -2.2 percent, while mixed-use occupations grew 6.9 percent.
The Yale Budget Lab’s October tracking finds the occupational mix not changing in ways that clearly align with AI; a Danish study (Humlum-Vestergaard) found chatbots produced no significant earnings or hours effects; and a CESifo paper we covered in September found US graduate unemployment “incredibly normal” through summer 2026. Three methodologies, one direction: mass displacement is not yet in the numbers.
Where the signal actually is: 17 losers, 20 winners, and one exact age band
The wiki’s occupation-level work is where it gets interesting for careers. Of the 135 exposed occupations, 17 shrank beyond the noise, shedding a combined 1.68 million jobs — customer service representatives (-8.7 percent) and computer programmers (-24 percent) among them. But 20 grew beyond the noise, adding 2.37 million — software developers +18.5 percent, project management specialists +30.6 percent, other computer occupations +24.4 percent. Exposure does not pick winners; the task mix inside the exposure does. That is the same pattern this site has tracked all year in the hiring data — AI postings surging 88 percent while the fastest-growing titles are deployment, security and QA roles, not research ones.
The young-worker finding is the sharpest the wiki has. In the most exposed quarter of occupations, 22-to-25-year-old employment grew 0.1 percent against 9.6 percent for young workers elsewhere — and older workers in those same exposed jobs grew 4.1 percent, unharmed. July-September’s 17.9-point gap against other young workers is the largest of the four quarters measured and the only one individually beyond the noise. It matches Stanford’s August 2026 revision — 19 percent relative underperformance for the same age band through reduced hiring, not layoffs — and Anthropic’s March research, which found the job-finding rate into exposed occupations down about 14 percent for the young, with no comparable hit to their elders. One cohort’s first door, narrowed — and this is exactly the mechanism NZIER described as “a perfect storm” for young New Zealanders.
What the trackers themselves say about their limits
The wiki is unusually honest about its error bars. Its caveats: the survey is large for groups but small for single occupations; the exposure scores date from 2023, measuring what AI could do, not what it did; and nothing here isolates cause. Employer statements — Amazon’s “fewer people doing some of the jobs,” Klarna’s assistant-did-the-work-of-700 claim followed by its re-hiring of humans — are classified as plans and claims, not outcomes. Challenger’s AI attribution (120,136 of 573,195 announced cuts through September, about 21 percent, in a year when total announced cuts fell 39 percent) appears hedged as employer-stated reasons. That is the discipline most AI-jobs coverage skips.
It also names what would change its mind, with dates: the October US jobs report on 6 November, Challenger’s October cut report on 5 November, and Census microdata through October — the open-questions page tracks each. If the young-worker gap keeps widening month over month, the “not yet” verdict has an expiry mechanism nobody else in the debate has committed to.
The New Zealand read
New Zealand’s data cannot yet run this exact test (occupation survey cells are too small), but the shape is visible in proxies: today’s Stats NZ filled-jobs release shows growth going to the 35-plus cohorts while under-30 employment contracts; SEEK’s gauge pairs record AI-skill demand with shrinking automatable-role ads; and MBIE’s own mapping put only 2 percent of NZ occupations at high automation risk. The international verdict — no economy-wide crisis, sharp entry-narrowing — is so far precisely New Zealand’s shape.
For young workers, the guidance in this data is concrete: the harm concentrates where firms hire juniors for exposed routine work, so the defensible moves are sideways into the occupations that grew beyond the noise — health, education, deployment and oversight roles — or up the task stack into the judgement-and-sign-off work the market is re-pricing upward (the review-economy shift). Entry is harder than in 2022. It is not closed, and the largest new categories — annotation, data centre operations — remain entry-friendly.
❓ FAQ
Has AI reduced overall employment? Not measurably. Jobs in the 135 most AI-exposed US occupations grew 2.8 percent in the year to September 2026 against 3.7 percent for the rest — within the survey’s margin of error, per the nimwiki tracker built on Census household data.
Which workers show a clear AI effect? Ages 22 to 25 in the most exposed occupations: employment essentially flat (+0.1 percent) against +9.6 percent growth for young workers in other occupations, with the gap widest (17.9 points) in the July-September 2026 quarter.
Which exposed jobs grew and which shrank? Software developers (+18.5 percent), project managers (+30.6 percent) and other computer occupations (+24.4 percent) grew beyond the noise; customer service reps (-8.7 percent) and programmers (-24 percent) shrank, among 17 declining occupations totalling 1.68 million fewer jobs.
What would change this verdict? The tracker names dated tests: the October Challenger cut report (5 November), the October jobs report (6 November) and Census microdata through October — if the young-worker gap keeps widening, “not yet” lapses.
Is the same pattern visible in New Zealand? In proxy form: NZ filled-jobs growth is concentrated in the 35-plus cohorts, SEEK’s automatable-role ads are shrinking, and NZIER documents youth unemployment at 25.3 percent against 3.5 percent for adults — the same entry-narrowing shape without an economy-wide decline.
🔍 THE BOTTOM LINE
The most careful public accounting of the AI-jobs question now agrees with the labour statisticians rather than the loudest voices on either side: no economy-wide displacement, a real and widening barrier for young entrants to exposed work, and a split inside the exposure itself where judgement-heavy roles grew while execution-heavy ones shrank. That is not a reason to stop worrying about the entry rung — it is a reason to aim interventions at the actual injury site: hiring channels for the young, not headcount across the whole economy. New Zealand’s numbers, thinner but pointed the same way, say the shape is already here.
📰 Sources
- nimwiki — AI and jobs: Which jobs is AI taking? (updated 26 October 2026)
- nimwiki — Young workers in exposed jobs (26 October 2026)
- Stanford Digital Economy Lab — Canaries in the Coal Mine (2026 revision)
- Yale Budget Lab — Tracking the Impact of AI on the Labor Market (October 2026)
- Hacker News — discussion of the tracker (28 October 2026)
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
