My Assessment of AI's Impact on the US Economy and Labor Market
B · Useful, limited Labour economist Gad Levanon sets out a sourced, chart-backed reading of US output, jobs and graduate data; informed and specific, but by his own account an interpretation without causal identification.
Summary based on the full text·Original: about 15 minutes to read
In short
A Labor Matters Substack post (1 October 2026) by Gad Levanon, a self-described former techno-sceptic. He argues the consensus that AI has not yet moved the US labour market is wrong. Output in office-heavy industries has grown while their employment fell for 36 months straight, young graduates are the one group doing badly in a low-unemployment market, and studies finding no effect rely on exposure scores that do not measure displacement. He expects the large office job cuts to arrive in the next recession.
Key numbers
- about 13 percentrise in real value added in finance, insurance, information and professional services (FIIPB), end-2022 to end-2025, while hours fell about 2 percent
- about 700,000jobs shed in FIIPB industries since April 2023
- 1.5 and 2.6 millionrange of missing white-collar jobs against normal FIIPB employment trends
- 36 percent ... Now it's 27 percentbachelor's over associate degree earnings premium at ages 22-24, then vs now
- 1.5 to 2 percentknowledge workers in jobs that would not exist without AI (Burning Glass Career History Dataset)
As reported by the source.
English·Web page·opens gadlevanon.substack.com
What it reports
- Underlying GDP growth about 3%; non-farm productivity likely above 2.5% a year in 2024-2026, more than double the pre-pandemic decade. FIIPB productivity about 4% a year; much of the rest of the economy near zero.
- Payrolls up about 1.3 million over 24 months to August 2026, nearly all in health care and social assistance. FIIPB employment below its year-earlier level for 36 straight months; every earlier run like that since 1990 came with a recession.
- Clerical work falling, professional work holding; he names HR specialists and insurance underwriters as professional occupations seeing automation. Judgment required is his best predictor of which white-collar occupations grew in 2023-2025.
- Unemployment for bachelor's holders aged 22-27 up about 1.5 points since late 2022, three times the rise for young non-graduates; the degree pay premium at 22-24 fell from 36% to 27%, lowest since at least 2008.
- Across 29 white-collar occupation groups, one common exposure index had almost no link to 2023-2025 employment change and the other pointed the wrong way.
- FIIPB quit and hire rates near their lows since 2001 ("job hugging"); the Bay Area's share of US market capitalisation rose from 17.5% in 2009 to about one-third.
How they got it
Author's analysis of public BLS, BEA, JOLTS and CPS series, his own trend estimates of "normal" FIIPB employment, a simple cross-occupation comparison against two published exposure indices, and Burning Glass Institute career-history data for AI-created roles. Methods are described in prose; no code or tables are published.
Limits
- No identification: he states that rates, pandemic overhiring, federal cuts and immigration hit the same period and cannot be separated.
- The missing-jobs range depends on his choice of counterfactual trend.
- The 29-group exposure test is a two-variable comparison, not a full model.
- The recession forecast is a judgment call by analogy to manufacturing in 2000-2010.
Why it matters for companies
The argument matches what many employers are doing quietly: holding headcount flat while output rises, and not refilling junior roles in anticipation of AI. Firms should know that the cheap decision now (leave the role unfilled) erodes the pipeline that produces experienced staff later. HR functions are named among the professional occupations already seeing automation. If the recession thesis is right, workforce plans built on "AI has not changed hiring yet" will be caught out by a sudden restructuring.
Summary based on the full text. Read the original for the full argument and data.Read the original on gadlevanon.substack.com
- Tags
- Workforce, Productivity