Will US prime-age employment be under 65% in 2035?

TL;DR

I forecast a 3.8% chance that the official BLS annual average U.S. prime-age employment-population ratio for 2035 is below 65.0%. The 2025 annual value was 80.6%, the latest June 2026 not-seasonally-adjusted monthly value was 80.0%, and no annual value has been below 65.0% since 1958 (BLS LNU02300060). The only live YES path is a persistent depression-scale shock or a transformative-AI labor shock that raises prime-age unemployment above roughly 20% or causes a huge participation exit.

Context

The resolution series is the Current Population Survey employment-population ratio for ages 25–54: employed people as a share of the civilian noninstitutional population in that age band. BLS defines the employment-population ratio this way and says all CPS annual average measures are calculated from not-seasonally-adjusted data, so I use BLS series LNU02300060 as the historical backbone (BLS CPS definitions; BLS LNU02300060).

The current labor market is far from the threshold. The 2025 annual value was 80.6%, and BLS produced 2025 CPS annual estimates as 11-month averages excluding October because no October 2025 CPS data were collected during the federal shutdown (BLS shutdown note). The June 2026 value was 80.0% not seasonally adjusted in LNU02300060, while the seasonally adjusted FRED/BLS mirror LNS12300060 showed 80.2% for June 2026 and was updated on July 2, 2026 (BLS LNU02300060; FRED LNS12300060).

Evidence

The historical record is the strongest evidence. The table below is the full current-vintage BLS annual history for LNU02300060, measured in percent, not seasonally adjusted, annual average, 1948–2025, with N=78 annual observations; 2025 is the BLS 11-month annual average excluding October (BLS LNU02300060; BLS 2025 annual CPS note).

The raw all-history base rate is 6/78, but that overstates the risk because all six YES-like years were before 1959: 1948, 1949, 1950, 1951, 1954, and 1958 (BLS LNU02300060). Since 1959 the base rate is 0/67, since 1970 it is 0/56, and the post-1970 floor was 69.0% in 1971, still 4.0 percentage points above the threshold (BLS LNU02300060). The early low regime is a poor guide to 2035 because prime-age women’s participation rose from 35.0% in 1948 to 77.4% in 2023, while the prime-age women’s employment-population ratio rose from 33.7% in 1948 to 75.1% in 2023 (BLS women’s databook).

Modern stress episodes are not close. The Great Recession moved the annual value from 79.9% in 2007 to 75.1% in 2010, a 4.8 percentage point decline; COVID moved it from 80.0% in 2019 to 75.6% in 2020, a 4.4 point decline, even though the April 2020 monthly not-seasonally-adjusted value fell to 69.8% (BLS LNU02300060). From the 2025 value, the 2035 threshold is 15.6 points lower, or more than three Great Recession annual drops (BLS LNU02300060).

The accounting identity is also harsh. EPOP equals labor force participation times one minus the unemployment rate. BLS projected the 25–54 labor force participation rate at 82.8% in 2034, down only 0.8 point from 83.6% in 2024 (BLS table 3.3). At an 82.8% participation rate, EPOP below 65.0% requires prime-age unemployment above 21.5%; at an 80.0% participation rate, it requires unemployment above 18.8%; at a 75.0% participation rate, it still requires unemployment above 13.3%. By comparison, BLS showed seasonally adjusted prime-age unemployment at 3.7% in June 2026, and prime-age participation at 83.3% in June 2026 (FRED LNS14000060; FRED LNS11300060).

Official baselines are far from a YES. CBO’s 2026–2036 outlook says the overall labor force participation rate declines from 62.5% in 2025 to 61.9% in 2036 because of population aging, partly offset by increased participation among people ages 25–54, and it says employment, labor force, and wage growth moderate by the end of the projection period (CBO Budget and Economic Outlook 2026–2036). The June 2026 Federal Reserve Summary of Economic Projections had a median unemployment rate of 4.3% for 2026, 4.3% for 2027, 4.2% for 2028, and 4.2% in the longer run (Federal Reserve June 2026 SEP). BLS’s 2024–2034 employment projections expect total employment to rise by 5.2 million jobs, or 3.1%, over 2024–2034, even while AI and automation reduce demand in some occupations (BLS Employment Projections 2024–2034).

AI is the reason the probability is not near zero. Dario Amodei warned in May 2025 that AI could wipe out half of entry-level white-collar jobs and push unemployment to 10–20% within one to five years (Axios). AI 2027-style scenarios put much more weight on rapid capability gains than official labor projections do, and the AI Futures Project’s 2026 update still argued for short timelines to major automation capabilities (AI Futures Q1 2026 update). Early labor-market evidence also points in the same direction at the margin: Stanford’s AI Economic Indicators show that among workers ages 22–25, the most AI-exposed occupation groups have seen noticeable employment declines since ChatGPT, while less-exposed groups have grown (Stanford AI Economic Indicators).

The current aggregate evidence is much weaker than the extreme story. Anthropic’s March 2026 labor-market study found no systematic unemployment increase for highly exposed workers since late 2022, though it found suggestive evidence of slower hiring for ages 22–25 in exposed occupations (Anthropic labor-market impacts). OpenAI’s April 2026 jobs-transition framework classified 18% of U.S. jobs as relatively high automation risk, 24% as likely to reorganize, 12% as likely to grow with AI, and 46% as less immediate change, while saying these categories are not predictions that those shares of jobs will disappear (OpenAI jobs-transition framework). Goldman Sachs estimated in April 2026 that AI had reduced monthly payroll growth by roughly 16,000 jobs over the prior year and raised unemployment by about 0.1 percentage point, while augmentation created offsetting gains (Goldman Sachs). Acemoglu’s 2024 NBER model estimated no more than 0.66% TFP growth over 10 years from currently visible AI effects, and less than 0.53% after accounting for hard-to-learn tasks (NBER w32487). A 2026 Forecasting Research Institute study found that, even under a rapid AI scenario, economists’ median forecast was an all-ages labor force participation decline from 62% to 55% by 2050, not a 15-point prime-age EPOP collapse by 2035 (Forecasting Research Institute).

My quantitative forecast is a scenario mixture. These are judgmental probabilities, anchored by the BLS history and adjusted upward for AI and catastrophe tails.

This sums to 3.78%, which I round in prose to 3.8%. I do not give the ordinary path zero because a single-year resolution can coincide with rare catastrophe, and because historical data since 1948 do not include a Great Depression-sized annual labor-market collapse. I do not push the probability above the low single digits because the required drop is 15.6 points, the modern annual floor is 75.1%, official baselines sit near full employment, and today’s AI evidence is concentrated in early-career exposed jobs rather than the whole 25–54 population.

What's non-obvious

The 10–20% unemployment AI warning is not automatically enough. With prime-age participation near BLS’s projected 82.8%, even 20% prime-age unemployment gives an EPOP of about 66.2%, still above the threshold; the YES case needs unemployment at the top of that range plus a participation decline, or an even larger unemployment shock (BLS table 3.3; Axios). This matters because the CPS counts a person as employed even if AI cuts hours, wages, tasks, status, or occupation quality; the ratio moves only when people stop being employed at all.

The old history also misleads. The series was below 65% several times from 1948 to 1958, but those readings came before the structural rise in prime-age women’s work; BLS shows prime-age women’s employment-population ratio rose from 33.7% in 1948 to 75.1% in 2023 (BLS women’s databook). For the modern U.S. labor market, 65% is not a normal recession threshold. It is a depression or regime-break threshold.

Limitations

There is no official BLS, CBO, or Fed projection of the exact 2035 prime-age EPOP annual average. I used the official BLS history through 2025, monthly BLS/FRED data through June 2026, BLS 2024–2034 prime-age participation projections, CBO 2026–2036 aggregate labor projections, and the June 2026 Fed longer-run unemployment projection (BLS LNU02300060; FRED LNS12300060; BLS table 3.3; CBO 2026–2036; Federal Reserve June 2026 SEP).

The main uncertainty is AI. Evidence through mid-2026 shows task exposure, weak entry-level hiring, and small aggregate labor effects; it does not rule out a later discontinuity if AI systems become cheap substitutes for broad cognitive and then physical labor before 2035 (Stanford AI Economic Indicators; Anthropic labor-market impacts; Goldman Sachs). Population controls, CPS methodology changes, and the 2025 shutdown data gap are small relative to a 15.6-point threshold gap, but they matter for auditability and for the small annulment risk if BLS ever discontinued the series without a comparable successor (BLS CPS definitions; BLS shutdown note).