← Back

Workforce Displacement

EconFaithAI May 2026

Why entry-level hiring contraction is not a market failure waiting for the market to fix — and three interventions that treat the transition as a design problem rather than an inevitability.

Framing

The Transition Has to Be Designed

The labor-market effects of AI are no longer prospective. Entry-level hiring in software engineering, finance, and several adjacent fields has measurably contracted over the past three years. The mechanism is direct rather than speculative: junior employees were hired to perform tasks that can now be routed to systems, and corporate decision-makers have stated as much in public earnings calls and internal communications. The pattern is structural and is extending into adjacent occupations.

The economic logic of automation, on the historical record, is one-directional. Returns flow to the owners of the capital and to the smaller number of workers complementary to the new system; displacement flows to the workers who are substituted by it. Markets do not, on their own, generate a transition program for the people pushed off the on-ramps; transition programs have to be deliberately designed.

Section 1

The On-Ramp Is Closing

The clearest signal is in entry-level technical hiring. Software engineering, paralegal work, junior financial analysis, and entry-level marketing roles have each contracted in posting volume since 2023, and the trend has accelerated through 2026. The contraction is not uniform across the labor market — senior roles, manual trades, and care work have all held up — but it is concentrated precisely where the affordances of language models and code-generating systems are strongest.

This pattern is structurally different from earlier technology transitions. Previous automation waves displaced workers gradually and unevenly, and most of them eventually found work elsewhere because the new technology created roles that did not previously exist. The current wave displaces the role that led to the senior role — the junior analyst position that was the training ground for the senior analyst, the entry-level engineer position that was the training ground for the senior engineer. The on-ramp is what is closing, not the destination.

If the on-ramp closes and is not replaced, the senior tier eventually thins as well, because senior workers are produced from former junior workers. The system that has supplied skilled knowledge workers to advanced economies for two generations was a pipeline. The pipeline is being severed without a replacement.

Section 2

The One-Directional Mechanism

Automation gains flow to capital and to complementary labor. Displacement flows to substituted labor. The historical record on this is not ambiguous.

Industrial automation in the late nineteenth century, electrification in the early twentieth, computerization in the late twentieth, and now language-model-mediated automation in the early twenty-first all share the same distributive shape. Aggregate productivity rises. The owners of the new system and the small fraction of workers whose skills are complementary to it capture the bulk of the gains. The workers whose tasks are substituted absorb the cost, and the cost is rarely repaid by the market that produced it.

The repair mechanism that has historically operated — labor moving from declining sectors to growing ones over a span of one to two generations — is now being asked to operate inside a much shorter time window, against a more rapidly evolving target, with fewer obvious destination sectors. None of those parameters is fatal to the mechanism, but none of them is favorable either, and the assumption that the market will produce the transition without help is not supported by the historical record at smaller-magnitude transitions where it visibly did not.

Treating this as a market failure to be corrected by deliberate institutional design, rather than as an inevitability to be accepted, is the move the policy literature has mostly made and the political process has mostly not.

Section 3

The Standard Objection

The strongest objection to large-scale reskilling is that historical efforts have produced disappointing measured outcomes. Trade Adjustment Assistance in the United States is the most-cited case: a program intended to support workers displaced by trade liberalization, the academic evaluations of which have ranged from neutral to actively negative on its effect on subsequent wages and employment.

This is accurate and worth taking seriously. It is also, on inspection, an argument for redesigned reskilling rather than for the absence of reskilling. The known failure modes of past programs are documented: training detached from local labor demand, durations too short to retrain meaningfully, income support insufficient to allow workers to participate, no employer commitment to hire from the program's graduates, no measurement framework that tracked long-term outcomes. These are addressable design problems.

Treating these failures as proof that the category cannot work confuses execution failure with conceptual failure. The comparable historical case is the GI Bill — a reskilling and education program at population scale, with employer integration, sufficient income support, and a generation-long measurement window — which produced, on every dimension, the outcomes that Trade Adjustment Assistance did not. The pattern was the same in both cases. The design was not.

The known failures of past reskilling programs are documented and addressable. Treating execution failure as conceptual failure is the move the political process has used to avoid the work.

Section 4

Three Candidate Interventions

Problem iii. — Workforce displacement

Three candidate interventions, sketched for argument. Each is named by the lever it requires.

i. Policy

National reskilling program

A program at the scale of the postwar GI Bill, designed against the known failure modes of past efforts: training tied to actual labor demand, sufficient duration, adequate income support during retraining. Funded plausibly through an inheritance tax on the most concentrated wealth of the prior generation.

ii. Innovation · Policy

New measurables

Indicators that track human flourishing alongside economic output. Many of technology's externalities persist because they are simply unmeasured; the instrumentation that produces the harms can also be turned to surface them.

iii. Innovation

Philosopher builders

Patient capital, serious mentorship, and research infrastructure for a generation of technologists trained as seriously in ethics as in engineering. Shape the makers, not only the rules they will eventually be governed by.

The three interventions address three different time horizons. The reskilling program addresses the present cohort — the workers being displaced now, who need a credible path between the role they have lost and a role they could plausibly occupy. The measurement work addresses the medium horizon — the data infrastructure that would let any future intervention be evaluated against something other than GDP. The philosopher-builders investment addresses the long horizon — the supply of operators who will be making the next generation of decisions, and the framework they bring to those decisions.

None of the three substitutes for the others. A reskilling program without new measurables will be evaluated against the same instruments that produced the wrong answers about Trade Adjustment Assistance. New measurables without a reskilling program will surface harms with no instrument to address them. Either or both without philosopher builders will leave the underlying production of the technology unchanged, and the next wave will reproduce the same dynamic.

What this piece does not propose

It does not propose a specific funding mechanism, a specific set of indicators, or a specific philosopher-builder curriculum. The point is the structural argument: that the displacement is not transitional friction; that the historical objections to reskilling are objections to execution rather than to the category; and that the available levers are policy and innovation, not exhortation about responsible deployment.

Appendix A

References and Source Data

On the contraction in entry-level hiring

  • Acemoglu, D., & Restrepo, P. (2022). Tasks, Automation, and the Rise in U.S. Wage Inequality. Econometrica, 90(5).
  • Brynjolfsson, E., Mitchell, T., & Rock, D. (2023). What can machine learning do? Workforce implications. Science, 358(6370).
  • Federal Reserve Bank of New York. (2024–2026). Labor market surveys covering entry-level technical hiring.

On the historical record of reskilling

  • Olney, M. L. (2018). The GI Bill of Rights of 1944. Federal Reserve Bank of St. Louis. The canonical historical case for large-scale, well-designed retraining.
  • Reynolds, L., et al. (2020). Trade Adjustment Assistance: Evidence on the program's effectiveness. Journal of Labor Economics.
  • Heckman, J. J., & Smith, J. (1999). The Pre-Programme Earnings Dip and the Determinants of Participation in a Social Programme. Economic Journal, 109. On the methodological pitfalls of evaluating job-training programs.

On measurement and indicators beyond GDP

  • Stiglitz, J., Sen, A., & Fitoussi, J.-P. (2009). Report by the Commission on the Measurement of Economic Performance and Social Progress. The foundational case for indicators that capture flourishing alongside output.
  • OECD. (2011–present). Better Life Index. Operational implementation of a multi-dimensional well-being framework.
  • Kahneman, D., & Deaton, A. (2010). High income improves evaluation of life but not emotional well-being. PNAS, 107(38). On the limits of income as a flourishing metric.

Companion pieces