Research

What the 1930s show

Young people were among the hardest hit by the Depression, and the damage did not end with the recovery. Research on workers who enter the labor market in a bad economy finds earnings penalties that persist for years, and work on people who come of age in recessions finds lasting differences in belief: a greater conviction that success is luck, and a different relationship with government and institutions.

What is notable is how the New Deal answered. The Civilian Conservation Corps enrolled roughly three million young men over its life, paying about $30 a month, most of it sent home to families. The National Youth Administration put well over a million students to part-time work to keep them in school. And as youth jobs vanished, schooling surged — the high-school graduation rate rose from under a third of young people in 1930 to about half by 1940. The thirties turned a jobs shortage into a family-income bridge and a skills investment at the same time. That dual design is the part worth copying.

What is different now

The 2026 data suggest the first measurable AI effect is not layoffs but a thinner front door. Employers hire fewer juniors; no one is dismissed, so there is no headline, no unemployment claim and no policy trigger. The young person who never gets the first job does not appear in the usual statistics at all.

A second, slower worry is the ladder problem: if the routine tasks juniors once learned on are automated, where do tomorrow’s seniors come from? This is a hypothesis, not a finding — but the cost of being wrong lands on a whole cohort. Equally, interest rates, post-pandemic over-hiring and policy uncertainty are plausible contributors to weak graduate hiring, so this paper treats AI as one contributor rather than the cause.

Mitigation options

• First-job hiring credits tied to trainingPayroll-tax credits for employers who hire and train new entrants into roles at risk of being automated away.Deadweight cost (subsidizing hires that would happen anyway) and employers swapping one junior for another.
• Paid apprenticeship at scaleExpand earn-while-you-learn pathways beyond the trades into the white-collar occupations where entry rungs are thinning.Quality control, employer capture of the training, and the slow build of a new institution.
• A modern service corpsCCC/NYA-style paid public-service entry roles with a training component, as a floor under first jobs.Cost, possible displacement of private hiring, and the stigma problem of any program seen as a last resort.
• Treat training like capital spendingGive employer training the same tax treatment as equipment, so that keeping and upskilling people is not penalized relative to buying software.Hard to target; benefits may flow to firms already training.

What the 1930s show

Agriculture was the Depression’s great technological displacement. Tractors were replacing mules, horses and hands across the South and the Plains. Then the Agricultural Adjustment Act of 1933 paid landowners to cut acreage. In the cotton South, many landlords reduced the number of tenants and sharecroppers on their land while the payments flowed to owners. Historians still argue over how much was the AAA and how much was mechanization and the collapse of crop prices, but the distributional result is not in dispute: the gains went to those who owned land and capital, and the costs to those with the least bargaining power. Sharecroppers organized — the Southern Tenant Farmers’ Union formed in Arkansas in 1934 — and many simply left.

Those who left met a second failure: relief rules tied to local residence. In 1936 the Los Angeles police sent officers to the California border to turn back migrants, the so-called “bum blockade.” The people Steinbeck wrote about and Dorothea Lange photographed were displaced by a technology and then penalized for moving.

The Rural Electrification Administration offers the opposite lesson. In 1935 only about one farm in ten had electric service, because private utilities judged rural lines unprofitable. The REA lent to cooperatives instead, and within about two decades most farms were connected. When a general-purpose technology diffuses on profit alone, it follows the profit; public design changed who benefited.

What is different now

AI exposure appears concentrated in clerical, customer-support and routine professional work, and unevenly across places. The “China shock” literature is the closest modern analogue and its message is sobering: local labor markets did not quickly re-equilibrate, many displaced workers did not move, and the effects persisted. Whether the gains from AI accrue mainly to capital owners and a few firms or are broadly shared is an open question, not a settled one.

Mitigation options

• Pay the displaced directlyAn “automation adjustment assistance” program, modeled on trade adjustment assistance, with income support and retraining going to affected workers.The trade version had low take-up and uneven training results; the AI version would need to learn from that.
• Wage insuranceTop up pay for workers who find a new job at lower wages, rewarding quick re-employment rather than prolonged search.Cost, gaming, and a tendency to favor older workers with long tenure.
• Place-based transition fundsRegional investment where exposure is concentrated, plus portable benefits so that moving is never penalized — the reverse of the 1930s residency rules.The perennial dilemma of propping up declining places versus helping people leave them.
• Broaden ownership of the gainsEmployee ownership, citizen capital accounts and public-fund proposals are all in debate as ways to share AI’s returns.Design complexity, investment risk and genuine disagreement over who should fund and control them.

What the 1930s show

For the first years of the Depression, relief was local, private and ad hoc. By 1932 many cities and counties had exhausted their funds. The federal government moved late and in steps: the Reconstruction Finance Corporation and loans to states in 1932, then direct federal relief grants under Harry Hopkins in 1933.

The sharpest demonstration of what speed can do came that winter. The Civil Works Administration put more than four million people on payrolls within about two months — proof that a federal jobs program can be stood up very fast when plans and administrators already exist. It was also wound down by spring. The permanent architecture came with the Social Security Act of 1935, which created federal–state unemployment insurance — but it excluded agricultural and domestic workers, occupations in which Black workers were heavily concentrated, a gap that took decades to close. Then, in 1936–37, support was pulled back too early and the economy fell into a second sharp recession. Frances Perkins was the architect of the insurance system; Marriner Eccles was among the officials who urged a renewal of federal spending once the slide began.

What is different now

Unemployment insurance is a genuine automatic stabilizer — but one built for layoffs of workers with a wage history. If AI’s labor effects arrive as hiring freezes and attrition, the people affected may never qualify: new graduates have no base-period earnings, gig and contract workers may fall outside, and a firm that quietly stops hiring creates no claim. Several tiered frameworks, from economists and from AI companies, now propose escalating responses keyed to unemployment thresholds, and one proposal would trip stabilizers if labor’s share of income drops sharply. None of this has been tested at scale, and the choice of trigger is the hard part.

Fiscal space matters too. The Trust’s Echoes of 1929 notes the IMF’s warning about abrupt increases in sovereign bond yields; a stabilizer financed in the middle of a bond-market scare is a different problem from one financed in calm.

Mitigation options

• Pre-authorized, trigger-based expansionsLonger or larger benefits that switch on automatically when pre-agreed indicators — including youth and entry-level employment — cross a line, rather than waiting for Congress. Extended Benefits already works this way for state unemployment rates.Choosing triggers that are timely, hard to game and not distorted by measurement lags.
• Eligibility modernizationAlternative base periods and coverage for new entrants, contract and gig workers.Higher cost and administrative complexity; risk of reducing the work-search incentive if poorly designed.
• Shelf-ready public employmentA standing authority and a pre-approved project list so that a CWA-scale response could start in weeks, not years — “build the binder before the winter.”Needs continuous upkeep; a plan nobody maintains is not a plan.
• Measure the hiring marginReal-time data on hiring and entry-level postings, not just layoffs: you cannot trigger on what you cannot see.Private data sources are uneven and can be misleading about causes.

What the 1930s show

Keynes, writing in 1930, popularized the term “technological unemployment” and predicted that within a century productivity would make something like a fifteen-hour week possible. He treated the intervening unemployment as a temporary maladjustment — a prediction that, so far, has proved right about the hours and optimistic about the transition.

The decade was, paradoxically, technologically rich: some economic historians rank the 1930s among the most productive stretches in American history. Output recovered faster than employment. Meanwhile, hours fell involuntarily and informally, and the government urged voluntary work-sharing. In 1933 the Senate passed Hugo Black’s thirty-hour-week bill, which would have spread the available work by law; it was sidelined by the National Industrial Recovery Act. What survived was the Fair Labor Standards Act of 1938: a 25-cent minimum wage and a maximum workweek that stepped down to forty hours, with overtime at time-and-a-half. The forty-hour week we still use is a Depression-era answer to the productivity-dividend question.

What is different now

Short-time work — where government tops up pay for reduced hours rather than paying for layoffs — was tested at scale in the 2008–09 recession, most famously in Germany. The research suggests it preserves jobs in a temporary downturn, but also carries costs when it props up jobs that will not come back. Structural change is different from a temporary shock, and subsidizing hours at firms that will not need those positions delays, rather than cushions, adjustment.

There is a longstanding economists’ objection to all work-sharing: the “lump of labor” fallacy, the idea that there is a fixed amount of work to divide. Its critics were largely right across past technology waves; whether AI changes that, because it targets cognitive tasks, is exactly what is disputed.

Mitigation options

• Short-time compensation as a shock absorberUse it for temporary adjustment, with sunset dates and training conditions.Labor hoarding and the risk of delaying a necessary shift.
• Review hours and overtime rulesThe 1938 overtime threshold shaped hiring incentives for decades; revisit whether it still does.Higher cost per hour may reduce hours worked rather than raise headcount.
• Bargained hour-sharing pilotsLet unions and employers negotiate productivity-gain sharing, including shorter weeks, with rigorous evaluation.Four-day-week trials to date involve self-selected firms and mixed results; national mandates are premature.
• Use freed time for retrainingPair reduced hours with paid learning, so the time is an investment rather than just a cut.Training quality, and the question of who pays for it.

What the 1930s show

Where institutions looked unresponsive, alternatives filled the space. The Townsend Plan promised $200 a month to the elderly, to be spent within the month. Huey Long’s Share Our Wealth clubs spread across the country in 1934. Father Coughlin, who began as a supporter of the New Deal, built a vast radio audience and drifted into hostility and antisemitism. They offered something visible to people who felt dispossessed, and some of what they offered was dangerous.

The administration’s answer was explicitly about dignity. Roosevelt and Hopkins preferred work relief to the “dole,” and the Works Progress Administration employed about 8.5 million people over its life, building roads, schools and airports and employing artists and writers as well. The program drew real criticism — for patronage, for make-work and for wages below private pay — and the Hatch Act of 1939 was passed in part to stop the political use of relief. It also fell short on inclusion: access was uneven, and it took A. Philip Randolph’s threatened March on Washington in 1941 to extract an executive order banning discrimination in defense industries.

What is different now

The legitimacy risk is not unemployment alone but the perception that the change was done to people, by unaccountable actors, with no one answerable. Media dynamics differ — the analogy between radio demagogues and algorithmic feeds is suggestive, but it is an analogy, not a finding. A further wrinkle: attributing layoffs to AI when other causes dominate (“AI-washing”) can distort both policy and public mood, in either direction.

There is a useful later precedent for worker voice. The 1960 West Coast longshore Mechanization and Modernization agreement traded work-rule flexibility for guaranteed earnings and benefits as containers arrived: workers accepted the technology because the terms of change were negotiated. The 1935 Wagner Act made that kind of bargaining possible.

Mitigation options

• Fast, visible, local public workA rapidly deployable service corps administered locally, with prevailing wages and strict anti-patronage rules.Crowding out private hiring, and the risk the program is perceived as make-work.
• Inclusion by designCoverage, anti-discrimination terms and appeals built in before launch — not retrofitted, as with Social Security’s exclusions.More complex administration, and more upfront political negotiation.
• Worker voice in AI deploymentNotice, consultation and negotiated transition terms where AI changes jobs, on the Wagner and longshore models.Uneven union coverage today; risk of slowing beneficial adoption.
• Honest, early-warning dataPublish displacement and hiring data plainly, and resist both over-promising and over-alarming.Data takes time to mature, and publication can itself move expectations.