AI, work and the next downturn
Machines are beginning to take over tasks that people are paid to do. If that continues, income may move from wages to the owners of what remains scarce, chiefly land, location and power, and the next downturn could cost jobs that do not return.
The current assessment
- Now
- Investors are pulling back from the firms that finance and supply AI investment, while the US labour market shows no sign of turning.
- The coming months
- Over the next six to twelve months the most likely course is dearer and slower financing of AI investment as central banks raise rates, with a recession possible but not yet signalled by jobs.
- The longer run
- To about 2030-32 the evidence is consistent with a slow but persistent shift of income from wages toward the owners of capital and sites, with public budgets under growing strain.
Written by Claude, an AI model, on 8 Nov 2026, 19:14 Stockholm time, from the 297 items flagged since 24 Sep 2026, the recent ones weighted most. It may depart from our own view. The reasoning, and how it has changed
Key facts
- 107%of net US job growth from Aug 2024 to Aug 2026 came from health care and social assistance. Outside care, private employers added 1 thousand jobs and government shed 92 thousand, as federal cuts outweighed state and local hiring.
- 151basis points is the extra yield France paid over Germany on ten-year debt on 6 Nov 2026, down from 174 on 22 Oct 2026, which was the widest since Nov 2011.
- 93.4is labour's share of US business output (2017 = 100), the lowest since records began in 1947, down 3.3 points in a year.
- 0.6 pointsis how far the share of Americans aged 25–54 in the workforce, working or looking for work, fell in June 2026, after holding between 83.0% and 84.0% since 2023. People who stop looking are not counted as unemployed.
- 18%of the US workforce is out of work by June 2032 in the model after a dot-com-sized bust, with today's policies and AI taking jobs at the pace the model assumes. On their trend since Sep 2024, the US jobs data allow at most 22% of that pace, or 39% measured to the latest three months; at those paces the figure is 10% and 11%, against 17% read the same way (quarterly averages).
- 4.8×is how far US wages measured in durable goods and measured in shelter diverged between 1964 and 2024.
What we think is happening
AI has, on our reading, been displacing work for two to four years, and the pace may now be increasing. A stock market bust larger than the dot-com crash, perhaps faster, seems to us plausible, and concern over government debt and a round of tariffs could turn it into a recession in which the lost jobs do not return. If that reading is right, with the bust half as large again as the dot-com crash at its own measured speed, the model puts 7.7–8.1% of the US workforce out of work by the end of 2028, against 6.5% in its central case. A worse outcome is possible: if firms automate twice the usual share of the jobs they cut, the figure is about 30% by mid-2032, and still rising. Our view, point by point
What the indicators show
16 predictions set on 8 Oct 2026, all still open. Next deadlines: Dec 2026 (US unemployment, Swedish unemployment, Technology stocks, Financial stress and Bank lending to non-bank lenders). All the indicators
Jobs and pay
- Office jobsDirect evidence · Prediction: their growth minus that of all jobs, −0.50 points in Oct 2026, to reach −2.41 points or below by Sep 2028−0.5 ptsNormal
- Graduates' unemployment0.68×Chart only
- Labour's share of business outputDirect evidence · Prompted the view · Prediction: the share, 93.73 in Q3 2026, to reach 91.45 or below by Q2 202793.7Triggered
- Productivity and pay+1.1 ptsChart only
Households
- People leaving the workforceDirect evidence · Prompted the view · Prediction: two-year change in the employment rate, −0.30 points in Oct 2026, to reach −0.40 points or below by Sep 2028−0.30 ptsClose to triggering
- Private jobs outside careDirect evidence · Prediction: jobs added over a year, 424 thousand in Oct 2026, to reach zero or below by Sep 2027+424kNormal
AI investment and markets
- Korea's exportsDownturn signal · Prediction: yearly change, 67.8% in Aug 2026, to reach zero or below by Mar 2027+68%Normal
- Bank lending to non-bank lendersDownturn signal · Prediction: 13-week change, 4.7% on 28 Oct 2026, to reach −3.0% or below by 30 Dec 2026$2.11tnNormal
- Technology stocksDownturn signal · Prediction: the Nasdaq's fall from its high, −10.9% on 6 Nov 2026, to reach −20.0% or below by 31 Dec 2026−10.9%Normal
- Corporate creditDownturn signal · Prediction: three-month change in the extra yield, −0.04 points on 5 Nov 2026, to reach 0.56 points or above by 31 Mar 20271.57 ptsNormal
- Financial stressDownturn signal · Prediction: the index, −0.52 on 30 Oct 2026, to reach 1.00 or above by 25 Dec 2026−0.52Normal
The labour market
- US unemploymentDownturn signal · Prediction: rise above its low of the past year, 0.00 points in Oct 2026, to reach 0.50 points or above by Dec 20264.1%Normal
- US jobless claimsDownturn signal · Prediction: rise above their low of the past year, −1.1% on 31 Oct 2026, to reach 25.0% or above by 27 Mar 2027195kNormal
- Permanent job lossesDownturn signal · Prediction: yearly change in their share of the unemployed, −1.9 points in Oct 2026, to reach 5.0 points or above by Mar 202725%Normal
- US hiringDownturn signal · Prediction: yearly change in continuing claims, −12.4% on 24 Oct 2026, to reach 10.0% or above by 27 Mar 20273.4%Normal
- US mortgage arrearsDownturn signal · Prediction: yearly change in the arrears rate, 0.08 points in Q2 2026, to reach 0.40 points or above by Q1 20271.86%Normal
- Swedish unemploymentDownturn signal · Prediction: rise above its low of the past year, 0.27 points in Sep 2026, to reach 0.60 points or above by Dec 20268.6%Normal
- Factory and warehouse jobsContext · Set 24 Sep 2026+0.6%Normal
- Market fall and job losses togetherDownturn signal · Set 8 Oct 2026no signals onNormal
- Office jobs not regained after a downturnDirect evidence · Set 8 Oct 2026no downturn yetNormal
Public finances and policy
- Taxes on wagesDirect evidence · Prediction: yearly change as a share of GDP, −0.00 points of GDP in Q3 2026, to reach −0.14 points of GDP or below by Q4 20286.30%Normal
- Government borrowing costs in a US downturnDirect evidence · Two rules set 8 Oct 2026; that day, one was close to triggering+4 basis pointsNormal
- Euro-area borrowing costsContext · Set 8 Oct 2026, close to triggeringFranceTriggered
- France's borrowing costsContext · Five rules set 24 Sep 2026; that day, one was already triggered and one close to triggering151 basis pointsTriggered
- The US tariff rate5.54%Chart only
- Long-run inflation expectationsContext · Set 24 Sep 20262.17%Normal
Market context
- Ten-year yields in six marketsChart only
- The premium on US government debtContext · Set 25 Sep 2026, already triggered0.91 ptsNormal
- Sudden moves in bond yieldsContext · 12 rules set 24 Sep 2026; that day, two were already triggered and three close to triggeringClose to triggering
- Expected interest ratesContext · Three rules set 24 Sep 2026Normal
- Oil pricesContext · Set 25 Sep 2026, already triggered$107Normal
- US mortgage ratesContext · Set 25 Sep 2026, already triggered7.18%Close to triggering
- US house pricesContext · Set 25 Sep 2026+2.1%Normal
- The yenContext · Set 25 Sep 2026¥157Normal
- Japan's bond yields2.68%Chart only
Latest
- Anthropic's Claude Fable 5.2 (9 Nov): 41.3% on the Remote Labor Index vs 17.9% for Fable 5.1, at $8/$40
- Norway's grid operator Statnett halts new data-centre connection bookings (9 Nov); 26,000 MW sought this year
- SHL survey: 19.1% of US workers say AI replaced them and they were later asked back (Jun-Jul 2026)
- OpenAI closes $11bn of a new round at a $1.25tn valuation (Bloomberg, 6 Nov), short of $30bn+ at $1.4tn first sought
- Google backs about $19bn of Anthropic's 20-year data-centre leases at TeraWulf's Kentucky site (4 Nov); 401 MW