I
Labor · AI · Demographics / 2026 → 2034
A working atlas of the American workforce in the era of AI.
Jobs going unfilled · 2034
9.83M
Workers AI displaces · net
+0.00M
Under no AI, baseline immigration, a 10% wage-growth cap. Move the levers and watch both move.
Two things are true at once: America is short millions of workers, and AI is set to displace millions more. In different jobs than the ones going unfilled. The economy can’t fill the work it has open and can’t use the people it sheds. Shortage and displacement don’t cancel; the pain just relocates.
no AI-specific shock
current policy
over 2024, by 2034
II
The rotation
baseline immigration · 10% wage-growth cap
AI doesn't shrink the problem — it rotates it.
As AI intensifies, the shortage eases (fewer cognitive workers needed) while additional unemployment climbs (displaced workers don't re-queue). Scroll to watch the two lines converge.
Shortage @ active wage-growth cap · Additional unemployment = AI-attributable (net of the no-shock baseline) @ rigid-wage floor · Incremental on top of today's ~4% structural rate.
III
The setup / two forces, same decade
For the first time, the workforce is aging out and a new technology arrives at once.
Two structural shifts are landing on the U.S. labor market in the same decade — one demographic and slow, one technological and sudden. Scroll the chart.
U.S. labor force participation rate
%, 1998–2026
Source: BLS / CPS via FRED — CIVPART, LNS11300060
1998 → today
Participation has slid for two decades.
The share of Americans working or looking for work fell from 67.1% in 1998 to 61.8% now. The pool everything else draws from has been shrinking.
the usual suspect, cleared
But prime-age workers didn't leave.
Among 25-to-54-year-olds, participation is essentially unchanged — 84.0% then, 83.9% now. The decline isn't people in their working years dropping out.
the slow shift
The gap is the workforce aging out.
Older cohorts are a growing share of the population and they participate less — even while working later than ever (55+ participation rose from 31% to 37%). That's the slow, demographic shift. AI is the fast one, and it lands on top.
−5.3pp
Total participation, 1998→2026 (67.1% → 61.8%), even as prime-age participation barely moved (84.0% → 83.9%).
~1.0opening / unemployed
Job openings per unemployed worker — above 1 again, after peaking near 2.0 in 2022. A structurally tight market.
25of 94
occupation groups are highly exposed to AI — the new shock landing on top of the demographic one.
How AI lands on the 94 occupation groups
Autor & Thompson, 2025
IV
The price of clearing
baseline immigration
To fill the jobs, wages must rise. Someone pays for that.
Clearing the shortage means paying more, and wage increases pass through to prices. AI relieves the average pressure by pushing some wages down — but the occupations families depend on still demand double-digit raises. Scroll to see the squeeze.
The price line is illustrative — drawn from the labor-cost literature, not the projection model. How we estimate this →
Unconstrained clearing wage (CSE) · CPI impact illustrative, cumulative over the decade · The relief under AI is paid for in displacement — see the rotation above.
V
What we can actually do
Two levers: move workers to the work, and move workers in.
Reskilling can route some of the displaced into the shortage, but not most of them, and never all. Of the workers displaced under transformative AI, about 49% have openings in their own occupation elsewhere, roughly 12% could move to a nearby field, and about 39% (≈3.8 million) would need deep retraining. Some shortage always remains.
Bucket analysis · preliminary
BetaHow much of the displacement can actually be absorbed?
Displaced workers, by scenario, sorted by absorption pathway. Same SOC = the same occupation has shortage somewhere (the worker doesn't need to reskill, just possibly move). Within-cluster = a different occupation in the same SOC-major (light reskilling). Cross-cluster = a destination cluster where CPS-observed transition rates clear a 3% threshold (medium reskilling). Stranded = the cube has no obvious destination for these workers — either no occupation is in shortage that matches, or the CPS pathway doesn't exist.
Transformative AI · baseline imm (headline)
10.55M displaced
Gradual AI · baseline imm
5.23M displaced
No AI · baseline imm
1.83M displaced
Transformative AI · no immigration
8.34M displaced
Under transformative AI, ~39% of displaced workers (≈3.8M) land in the stranded bucket. The model has no obvious destination for them at this scenario: their occupation is in surplus, the within-cluster pool is too small, and the CPS doesn't show meaningful transition rates into the clusters that ARE in shortage. That's the bucket that demands deep reskilling — or it's a bucket the labor market doesn't absorb.
Lower immigration shrinks the stranded share (39% → 25% at no-new-immigration), not because fewer people are displaced (though that helps), but because tighter labor supply means more shortage in the SAME occupations as the displaced — bucket 1 (mobility-only absorption) rises from 48% to 67%.
Where stranding concentrates (transformative AI, baseline imm)
Eight occupation clusters carry most of the stranded pool.
| SOC-major | Displaced | Stranded | Stranded share |
|---|---|---|---|
| 15 Computer & Mathematical | 398K | 383K | 96% |
| 43 Office & Admin Support | 1.45M | 1.37M | 95% |
| 13 Business & Financial Operations | 785K | 736K | 94% |
| 27 Arts, Design, Media | 254K | 199K | 78% |
| 11 Management | 1.72M | 1.17M | 68% |
| 17 Architecture & Engineering | 256K | 158K | 62% |
| 29 Healthcare Practitioners | 435K | 254K | 59% |
| 33 Protective Service | 150K | 63K | 42% |
Office Admin, Business & Financial Operations, and Computer & Mathematical sit at the top of the stranded list because the model says those occupations are in surplus everywhere under transformative AI — and the empirical CPS doesn't show meaningful transition rates from these clusters into Healthcare, Personal Care, or Construction (the clusters that ARE in shortage). The policy implication isn't subtle: reskilling pipelines targeting these three clusters specifically would absorb a disproportionate share of the stranded pool.
Methodology
For each AI scenario, displaced workers (post-CPS re-queue, held at the rigid wage floor) are matched against the shortage profile at the same scenario via a sequential SOC-level allocation. Largest displaced SOCs go first.
Bucket 1 absorbs into the same SOC-minor wherever it has shortage. Bucket 2a spills over into other SOC-minors in the same SOC-major (within-cluster). Bucket 2b attempts cross-cluster moves, but capped at d × Σ p where p is the CPS empirical transition rate and only destination clusters with p ≥ 3% are eligible. Bucket 3 is the residual.
CPS transition rates come from BW-LaborShortages' 22-group transition matrix (annual-equivalent). The 3% threshold is robust within ±10% (sensitivity sweep in script output: B3 ranges 3.5M–4.1M across τ ∈ [0.5%, 8%]).
Framework adapted from Escobari, Seyal & Daboin Contreras (2021), Moving Up, Brookings — which uses the same cluster-then-CPS logic on observed worker transitions. Our cube provides the volumes; the matching logic is theirs.
Caveat. This is Session 2 of three. Per-metro matching (which would separate "same SOC, same metro" from "same SOC, must relocate") is deferred. So bucket 1 currently bundles within-metro absorption with cross-metro relocation. Roughly half of bucket 1 likely requires moving — expect that to be a major policy point once Session 3 lands.
The second lever is migration, and it cuts both ways. Immigrants arrive already trained (42% of home-health aides, about 28% of physicians), filling shortage roles in one to three years against the roughly twenty it takes to grow a worker from a birth cohort. But cutting immigration eases AI-driven displacement while it widens the shortage: the two headline numbers move in opposite directions under the same policy.
VI
Elsewhere on the atlas
Three depths
Three ways to go deeper.
01
Atlas
Every occupation, every metro, every cell of the cube.
Projected gaps by occupation and metro under every combination of AI, immigration, and wage adjustment.
Open →
02
Policy
Reskilling and migration, scenario by scenario.
Who can be reskilled into the shortage, who stays stranded, and how migration policy moves both numbers at once.
Open →
03
Pulse
What the country is already saying about its workforce.
10,681 full-text articles on labor shortages (2023–2026), and whether the coverage points at the real gap.
Open →