Can America retrain workers before AI leaves them behind?

Section: United States Edition: 2026-08-08 Edition

One-line thesis

The Economist's specific-case answer to the fiscal-Leader's "job disruption → unemployment spending" offset: America has never had a credible worker-retraining system at scale, and the AI shock is arriving with no better plan. Goldman Sachs estimates 10 million jobs could be displaced over the next decade. Federal training spend is $170m on 39,000 people in 2023; Trump's One Big Beautiful Bill Act extension of Pell grants adds only ~100,000 additional recipients/year by 2034 at $2,200 each; the largest philanthropic ask so far is OpenAI's $250m + Anthropic's $200m vs OpenAI's own ~$600bn compute spend through 2030 (~800× the training pledge). The scale mismatch is the story.

Key numbers

The scale of coming displacement:

~10m jobs could be displaced over the next decade per Goldman Sachs.
~1 in 5 workers expects AI to eliminate their job within five years — the fear is already shaping US politics regardless of when/whether displacement lands.
The shape differs from prior shocks: entry-level white-collar + routine office work most exposed; losses via less hiring + layoffs, not factory closures.

Historical baseline — Trade Adjustment Assistance (TAA), 2000-2011:

Nearly 6m factory jobs lost 2000-2011 (~1m from Chinese import competition).
TAA covered ~160,000 workers/year while manufacturing shed ~500,000/year — a 32% coverage rate.
Only ~40,000 entered training/year (cumbersome applications + long waits).
Trained workers earned ~$50,000 more over the following decade than comparable non-trained workers — the programme worked for those who reached it.
TAA closed to new workers in 2022.

The current federal-support baseline:

In 2023, the main federal training scheme for displaced workers spent $170m on ~39,000 people — about $4,350 per participant.
Community colleges: ~$25bn public funding for workforce programmes, serving ~8m people/year.
~40% of American high-school graduates enrol in a four-year university — the assumption baseline the education system was designed around.
~800,000 registered apprentices in America (~half Germany's total with 4× the population); concentrated in construction/skilled trades.

The lab-philanthropy scale mismatch:

Anthropic: $200m committed for research + programmes to help workers navigate AI disruption.
OpenAI Foundation: $250m committed.
OpenAI plans to spend ~$600bn ($200bn × 3, ~3,000×) on computing power through 2030. The Economist's specifically-quoted comparison: OpenAI's compute-through-2030 spend is ~3,000× its worker-retraining pledge.
Trump's One Big Beautiful Bill Act: extended federal Pell grants to approved career programmes lasting 8-15 weeks. CBO expects the change to reach only ~100,000 additional recipients per year by 2034, with grants averaging just $2,200.

The new federal-response actor: RAISE US

RAISE US — nonprofit co-founded by Gina Raimondo (Joe Biden's Commerce Secretary + former Rhode Island Governor) and Eric Holcomb (former Republican Governor of Indiana).
Donors include: Anthropic, OpenAI, General Motors, Bank of America.
Pilots in 4 states. In Maryland, built on a state scheme that pays young workers to spend a year with non-profits — adding coaching + training + pathways into long-term health-care and education jobs.
The specific enterprise-scale template the piece uses: Salesforce — maps employees' skills to possible internal roles, recommends training, reimbursement can cover formal qualifications. "It is fitting that Salesforce, itself being reshaped by AI, is testing ways to help its staff evolve. But such internal platforms remain rare."

Load-bearing quotes

Raimondo: "Our whole system is predicated on funding based on attendance, not outcomes." — the diagnosis of why the $170m/39,000-workers baseline underperforms.
Raimondo (RAISE US framing): "If we don't get this right, we won't lead the world in AI. There will be regulatory backlash, which is bad for the companies." — the self-interested-industry-support logic for firms funding retraining.
Ned Lamont (Governor of Connecticut): "Frankly, I need help from Anthropic and the AI companies to stimulate our imagination."
Lee Lilley (North Carolina Commerce Secretary): "There is a cost to people's egos. It is not realistic to take someone with a 20-year career in a white-collar setting and ask, 'Would you like to become a welder?'" — the skill-adjacency + identity-preservation constraint on retraining program design.

The Kimberly Brady vignette

The Economist opens with Kimberly Brady, 53, California, psychology degree, Costco electronics/jewellery sales job. After 620 unsuccessful applications for customer-service/sales roles (fields "increasingly reshaped by AI"), she decided to retrain for Costco's optical department. No formal programme available — she spent 6 months teaching herself opticianry from textbooks and YouTube, passed two national board exams, paid thousands of dollars for two licences. Without hands-on experience, "they still wouldn't hire me." Eventually moved into optical through an ad-hoc accommodation, not a scaled programme.

The vignette is the anti-datum the piece is designed around: the individual is motivated, has degrees, has invested cash and 6 months — and the system's answer is still an ad-hoc accommodation rather than a designed transition pathway.

Two workforce-tier reframes worth carrying

"Sectoral programmes" — bring employers + training providers together to design courses around actual vacancies. Four randomised trials in health care, IT support, manufacturing found lasting earnings gains of 11-40%. The Economist's endorsement.

The incumbent-worker challenge is separate from the entry-worker one. The Salesforce template is incumbent-focused; RAISE US is entry-focused. The two require different scaffolding — the piece is unusually clear that large white-collar employers helping staff learn how to use AI is common; helping them into a new role entirely is rare because "no one yet knows what they should be trained for."

Why this piece matters for the vault

This is the specific-case US-side complement to the same-edition cover-package Leaders:

Governments Are Making a Dangerous Bet on the AI Boom (Economist) — Brookings modelling explicitly names "AI-induced unemployment spending" as one factor that could "more than halve" AI's positive fiscal impact. This piece describes the specific mechanism that has to work to close that offset — and it isn't yet working. The fiscal Leader is long the retraining system; this piece is the audit.
Why AI Is a Risk to Communist China (Economist) — the Chinese Leader's "even Denmark and Singapore struggle" line applies symmetrically here. This piece confirms the American retraining record is worse than either.
FOBO (Fear of Becoming Obsolete) — the rational end of FOBO gets first-order support from this piece. The "1 in 5 expects AI to eliminate their job within five years" datapoint pairs with the vault's already-tracked Anthropic McCrory finding that workers in Claude-automated roles express greater fear of losing their jobs even though displacement hasn't materialized — see Why AI Hasnt Increased Unemployment According to Anthropic (AI Daily Brief). The retraining-system-absence documented here explains why the fear stays rational at the population level.
AI Layoff Trap — Tsoukalas & Falk's game-theoretic model has its policy-design partner here. Their fix is reprice the replace-vs-augment margin (replacement tax / retention subsidies); Raimondo's ask is industry-funded pre-emptive retraining. Different levers, same problem.
The Upskilling Bill — Whittemore's enterprise-side coinage gets its national-fiscal-scale mirror here. The scale of the bill is real; the current answers are ~800× under-scale.

New entity/concept pages this Leader anchors

Gina Raimondo — former Biden Commerce Secretary + Rhode Island Governor + RAISE US co-founder. First substantive vault touch; the "funding based on attendance not outcomes" diagnosis and the industry-self-interest → retraining logic framing are quotable and durable.
RAISE US — new entity page. Named-nonprofit worth watching as the first coordinated response to the retraining scale gap.
Salesforce — no vault page yet; this is the first substantive touch. Its internal skills-mapping platform is a first-order enterprise-IT-relevant example of what a mature AI-transition-support system looks like. Worth an entity page.

Cross-references

Governments Are Making a Dangerous Bet on the AI Boom (Economist) — the fiscal-Leader partner
Why AI Is a Risk to Communist China (Economist) — the paired-Leader on the Chinese-side retraining record
How AI Is Breaking the British State (Economist) — the state-capacity-side Leader whose "institutional-legitimacy-erosion" thread pairs directly with the retraining-gap thread
FOBO (Fear of Becoming Obsolete) — the fear-side of the same story
AI Layoff Trap — the game-theoretic mechanism this piece's policy discussion sits inside
The Upskilling Bill — Whittemore's enterprise-scale coinage, mirrored here at national scale
Anthropic · OpenAI — $200m + $250m commitments; 3,000× compute-vs-retraining scale mismatch
RAISE US — new entity
Gina Raimondo — new entity
Salesforce — new entity
Anthropic Economic Index — the automated-user-optimism finding sits under this scale story; individual optimism does not solve aggregate-labour-market transition
Indias IT Sector Is Surviving Artificial Intelligence (Economist) · The Philippines Big Offshoring Industry Is Growing Despite AI (Economist) — the same-edition services-labour paired pieces
Hourglass Organization · Deskilling Trap (Juniors) — the entry-level rung problem this piece confirms is happening

Source

Original: source
URL: https://www.economist.com/united-states/2026/08/02/can-america-retrain-workers-before-ai-leaves-them-behind
Print headline: "Retraining America"