Read Further · checked 26 Oct 2026
The evidence, annotated
Each source measures something different. Read them as different instruments pointed at the same moving target. Publication dates are given where the publisher states them.
ILO & NASK — Generative AI and Jobs: A Refined Global Index of Occupational Exposure
ILO Working Paper 140, May 2025. Scores nearly 30,000 tasks using AI-assisted scoring, a worker survey and expert validation, grouped into four exposure gradients. Measures: potential exposure. Does not measure: job losses or adoption. Notable: one in four workers in occupations with some exposure; higher exposure for women and in high-income countries.
World Economic Forum — The Future of Jobs Report 2025 (key findings)
January 2025. A survey of 1,000+ employers across 55 economies. Measures: what employers expect for 2025–2030 (170 million jobs created, 92 million displaced). Does not measure: realised outcomes. Employer expectations can be wrong in either direction. Full PDF.
McKinsey — The State of AI (survey series)
2025 edition published 5 Nov 2025; 2026 edition “On the road to ROI”, 25 Aug 2026. Surveys of organisational respondents (1,993 in 2025). Measures: reported use, scaling, expected and reported workforce effects. Caveat: self-reported by organisations, not workers. The 3 Sept 2026 chart compares 2025 expectations (32% anticipated reductions) with 2026 reports (14%).
Brynjolfsson, Chandar & Chen — Canaries in the Coal Mine?
Stanford Digital Economy Lab; first Aug 2025, revised 12 Aug 2026. US payroll data (ADP). Measures: employment trends by age and AI exposure. Findings framed by the authors as descriptive, not causal: no economy-wide displacement; young workers in exposed occupations lag peers, mainly via hiring. Contested: critics such as Antonio Casilli argue alternative explanations deserve more weight. US data may not transfer to India.
NITI Aayog — Roadmap for Job Creation in the AI Economy
October 2025, with NASSCOM and BCG. Focused on India's technology-services sector; presents divergent scenarios (losses up to 1.5 million, or creation up to 4 million) contingent on policy. Caveat: a policy roadmap with industry input, not an independent forecast. A companion roadmap addresses AI for India's informal workforce.
Anthropic Economic Index
Ongoing reports since Feb 2025; recent editions Jan, Mar and Jun 2026. Measures: how people actually use one AI assistant across tasks, including automation-like versus augmentation-like use. Caveat: published by an AI developer and limited to its own product's users. (This atlas was built with help from an AI model made by the same company; we flag the connection.)
Future of Human Work — a Sahilverse companion
A qualitative work-design audit and essay on dignity, autonomy and AI augmentation. Editorial, not research.
How to read these together
- Exposure studies (ILO) tell you where change is possible.
- Employer surveys (WEF, McKinsey) tell you what organisations expect or report.
- Payroll studies (Stanford) tell you what employment did — in one country.
- Usage data (Anthropic) tells you how a tool is used — by its own users.
- Policy roadmaps (NITI Aayog) tell you what a government intends and fears.