Search the Practicum

Type two letters · ↑↓ select · Enter open · Esc close

Type at least two letters.

AI and the labor market: what the data says

This is a live Practicum page for Volume 2. Verified: 2026-08-18. The figures become outdated, so check the date and the primary source. It supports Chapter 8 of Volume 2, "The New Economy: Labor Markets and Platforms."

Discipline: data is not opinion. Forecasts and surveys carry clear labels such as "analyst forecast" or "verify." This is a measured view, not a claim that AI has replaced everyone. Whether a company augments people or replaces them is a choice, not a fixed property of the technology.

Use these points as reference markers, not prophecies. They show where demand for labor is moving, who receives the first gains, and what that may mean for you.

Global reference points

  • WEF Future of Jobs 2025, published Jan 2025, surveyed more than 1,000 employers covering about 14 million workers in 55 economies. It projects 170 million new roles and 92 million displaced roles by 2030, for a net gain of 78 million, or about 7%. It estimates structural labor-market churn at 22%. See the source below.
  • ILO research on occupational exposure to generative AI points to task transformation more often than replacement of an entire occupation. In the ILO estimate, AI is more likely to augment most jobs than automate them completely. See the source below.
  • Entry work is changing first. Junior tasks often produce standard output that another person can check. The value of framing, verification, and responsibility rises. This pattern runs through Chapters 4, 5, and 7 of Volume 1.

The entry-level squeeze: new data from 2026

Chapter 8 argues that the profession does not disappear first. Its entry route breaks first. Stanford data now gives us a measurable case. Keep data separate from opinion:

  • Stanford Digital Economy Lab, "Canaries in the Coal Mine?" by Brynjolfsson, Chandar, and Chen was published in Aug 2025 and updated in Nov 2025. Using data from ADP, the largest US payroll processor, it found that employment among early-career workers aged 22 to 25 fell by about 13% from late 2022 in occupations with the highest AI exposure, including software development and support. Employment among older workers in the same roles was stable or rising. See the source below.
  • A 2026 update reports that employment for workers aged 22 to 25 continued to contract at about 3.8% per year by Apr 2026. Its contribution to total unemployment was still small, about 0.1 percentage point. This looks like a redistribution of entry routes, not a labor-market collapse. Regional Federal Reserve reviews, including Dallas Fed work from 2026, point in the same direction. Verify the current primary source.

The same squeeze outside the US, and what the aggregate data does not show

The Stanford work is US payroll data. It is worth asking whether the pattern travels, and the honest answer has two halves.

The aggregate European numbers show no collapse. Eurostat put euro area youth unemployment at 14.8% in June 2026, down from 14.9% in May, with 2.351 million people under 25 unemployed. Overall euro area unemployment was 6.3%. Those are not the numbers of a labour market being dismantled by software. If you have read that AI is destroying entry-level work across Europe, the official series does not carry that claim.

At the same time, sector-level reporting outside the US does describe a squeeze on graduate hiring in tech specifically, and national bodies have started naming AI adoption as one factor among several in youth labour-market entry. Both things can be true: a real narrowing in particular occupations, invisible in a whole-economy average.

Do not quote the percentages you see circulating

Figures for the drop in entry-level or graduate tech hiring vary enormously between sources, from roughly 45% to over 70%, because they measure different things: job postings versus hires, tech versus all sectors, and different baseline years. Several widely shared numbers trace back to secondary write-ups rather than a dataset. This page deliberately gives you the shape and the two figures that come straight from a statistical agency. If you need a number for a decision, go to the primary series, and state which one you used.

Business angle: when the junior step shrinks, a company loses its usual training ground for future specialists. Where will your mid-level staff come from in three years? That question affects hiring and training now. Use Who owns your ladder? to diagnose dependence on hiring platforms.

Historical calibration: Engels' pause

One fact from economic history helps with calibration. Robert C. Allen reported in a peer-reviewed 2009 paper in Explorations in Economic History that British output per worker rose by about 46% from 1780 to 1840, while real wages rose by only about 12%. Between 1840 and 1900, wages caught up: output rose 90% and wages 123%. The measured lesson is that technology's gains can bypass workers for a generation before reaching them. This is not a verdict. It shows why staying useful depends on bargaining power as well as skill.

Compute as the new capital: who gains first

When intelligence becomes cheaper, owners of scarce inputs gain first. Those inputs include chips, data centers, energy, models, and access to demand. The scale figures are forecasts, not facts. Analysts estimate AI capital spending in the trillions of dollars between 2026 and 2031. One Goldman Sachs estimate is about $7.6 trillion. The bottleneck may be the power grid, not money. Read who owns the rails for more detail.

Russian reference points to verify

Russian figures are secondary and need to be checked against their primary sources. Treat them as leads, not facts:

  • Yakov and Partners / Yandex report that the share of Russian companies using generative AI in at least one function rose from about 54% in 2024 to about 71% in 2025. Marketing and sales were the most common functions. See the source below.
  • VCIOM tracks awareness and use of AI among people in Russia. Check the figure and year in the primary source.
  • GetPayAll published estimates of AI service use, around 68% and 84% in May 2025. This is a secondary estimate, so verify it.
  • Sber said in 2025 that the ability to work with AI was a requirement for employees and candidates. See what to learn now.

What this means for you

Waiting for labor to catch up over a generation is not a strategy. Build the things that give you bargaining power now: ownership of the outcome, context, trust, and distribution. Start with what to learn now. For the business view, use the Volume 2 workbook.

Sources (verified 2026-08-18)

Version: 2026-07. What changed in R2: WEF Future of Jobs 2025 and Stanford "Canaries" now include specific figures and links. ILO and Yakov and Partners, including the 54% to 71% change, were added. Open verification markers were replaced with clear caveats such as "analyst forecast" and "verify" because some underlying items are secondary by nature. Review once a quarter and when a major report is published. Do not present a forecast as a settled fact.

AI and the labor market: what the data says