News & Research Updates

Three years after ChatGPT, UK Annual Population Survey data covering 412 occupations shows no aggregate displacement signal. Wages in high-exposure occupations have grown more slowly since 2019 — but the trend predates ChatGPT. Programmers and finance analysts continue to grow; admin and clerical roles have contracted. Adoption is concentrated: ~20% of tasks account for the vast majority of usage.

46% of London's workforce (~2.4M people) are in roles where GenAI could automate a share of their tasks — substantially higher than the UK average of 38%. Over 300,000 are in highest-exposure routine administrative roles. Worker- and business-reported AI adoption has roughly doubled in two years to 26–35%. Mayor Khan announces a London AI and Jobs Taskforce.

Theoretical paper showing that competitive demand externalities trap rational firms in an automation arms race that exceeds the collectively optimal level — even when firms can foresee the demand cliff. Capital income taxes, worker equity participation, UBI, upskilling, and Coasian bargaining all fail to fix it; only a Pigouvian automation tax restores the cooperative optimum.

Uses Sweden's natural experiment (Riksbank rate hike preceded ChatGPT by seven months) to disentangle monetary tightening from AI. The aggregate posting decline aligns with monetary policy, not AI. But within employers: 22–25 year olds in high-AI-exposure occupations saw employment fall 5.5% by early 2025, while 50+ rose 1.3%. First population-register evidence of the 'canaries' effect outside the US.

Introduces a new measure (Observed Exposure) combining theoretical LLM capability with real-world Claude usage. Computer Programmers 75% covered, Customer Service Reps next, Data Entry Keyers 67%. No systematic unemployment rise for highly exposed since late 2022, but suggestive ~14% drop in job-finding rate for 22–25 year olds in exposed occupations.

Survey of ~750 US CFOs: more than half of firms have invested in AI, with productivity gains concentrated in high-skill services and finance (≈0.8% labor productivity growth in 2025, expected >2% in 2026). Aggregate employment expected to fall <0.4% in 2026 — but composition shifts: routine clerical down >2pp over three years, with offsetting growth in skilled-technical roles.

UK Government's authoritative assessment. Five judgements: AI capabilities improving rapidly (task time horizon doubling every 7 months); ~70% of UK workers in AI-exposed occupations (vs ~60% advanced-economy avg); UK productivity gains 0.4–1.2pp annually possible; hiring falling faster in exposed occupations (UK ads -38% high-exposure vs -21% low-exposure 2022–2025); 16–24 year olds in computer programming -44% in a single year.

Causal study of 12,000+ EU firms + 800 US firms. AI adoption increases labor productivity by 4%, with gains driven by capital deepening rather than employment displacement — AI augments worker output, does not replace labour in the short run. Adoption stratified: large firms 45% vs small firms 24%; Sweden 52%, US 34%, Romania 22%. AI-adopting firms pay higher wages.

Cross-country analysis using Lightcast vacancy data (Brazil, Denmark, Germany, South Africa, UK, US). About 1 in 10 advanced-economy postings demand at least one new skill (3–3.4% wage premium in US/UK). AI-related skills boost wages but do NOT raise overall employment — and for high-AI-exposure low-complementarity occupations, employment levels are 3.6% lower in regions with higher AI-skills demand 5 years on. Reinforces job polarisation, potentially shrinking the middle class.