الجدول الزمني لسياسات الذكاء الاصطناعي والعمل

The Rebundling of Freelance Work in the AI Age

arXiv

This arXiv paper looks at how freelance work is changing in the AI era. The authors compare the activities people carry out in ChatGPT with the work clients request on Upwork, an online freelance platform, and examine how clients combine tasks into projects. Activities that are more common in ChatGPT tend to lose marketplace share on Upwork, and less common ones tend to gain share. Within AI-related work, demand is becoming more diverse, and job postings combine tasks that were rarely requested together before generative AI.

East Asia and Pacific Economic Update, October 2026 : Riding the AI Wave (Executive Summary, Chinese)

World Bank

This is the Chinese-language executive summary of the World Bank's East Asia and Pacific Economic Update for October 2026, "Riding the AI Wave". The Chinese summary was posted on 12 November 2026; the English report came out in October. It says a global surge in AI-related activity is lifting growth in economies that take part in AI supply chains, while high energy costs weigh on others, especially Pacific Island economies. AI use in the region is lower than in advanced economies, and firms name cost, lack of relevant knowledge, and security and privacy concerns as barriers. Firms' demand for AI-related skills is often paired with analytical and soft skills.

Gondola doubles revenue as its team builds more with Codex

OpenAI

An OpenAI customer story about Gondola, a company that helps creatives get credit and get hired. According to the title, Gondola doubled its revenue as its team built more with Codex. The feed summary says Codex raised the lead engineer's output by 50% and that ChatGPT simplifies daily work. This entry is based on the feed title and summary only; the full article was not read.

ChatGPT Plus for veterans and servicemembers

OpenAI

An OpenAI post on its free year of ChatGPT Plus for veterans and transitioning service members. According to the excerpt, more than 20,000 of them have claimed the offer, and the post describes how they use it to prepare for civilian careers. The figure is published by the vendor; the full text could not be read, and only the published excerpt is summarized here.

Singapore Airlines elevates travel & work with OpenAI technology

OpenAI

An OpenAI customer story on Singapore Airlines. According to the excerpt, the airline uses OpenAI models and the API for customer support, travel discovery and employee productivity, and customer satisfaction (CSAT) for its Kris chatbot doubled. Single-company figure published by the vendor; the full text could not be read, and only the published excerpt is summarized here.

Work at the Frontier: AI’s effect on freelance work

OpenAI

Research from OpenAI and Upwork on how AI is affecting freelance work. According to the announcement, it finds shifting demand for freelance work, new skill combinations, and larger budgets and more hours for AI-related projects.

TaskNet: Building Occupational Task Inventories from Worker Interviews at Scale

NBER

NBER Working Paper w35987 (Wang, Haupt, Robinson, Lewis, Brynjolfsson, Yang) introduces the Stanford Task Network (TaskNet), an occupational task taxonomy built from worker interviews rather than analyst-written descriptions. A mixed-initiative system interviewed 132 workers in six occupations and distilled 2,726 reports into 202 task statements, all accepted by O*NET Center analysts, versus roughly 65% acceptance for write-in suggestions under the current O*NET process. Against O*NET 31.0 the inventory covers 75% of 118 published tasks, adds 79 new and 58 refined tasks, and shifts AI exposure estimates in both directions, raising them for software developers and lawyers.

Cult.fit builds operational muscle with Codex

OpenAI

An OpenAI customer story on Cult.fit, an India-based fitness company. According to the excerpt, teams across functions are using Codex to build their own tools, move faster, and change how work gets done. The full text could not be read; only the published excerpt is summarized here.

Australian universities are creating AI-native graduates

OpenAI

An OpenAI customer story on Australian universities. According to the excerpt, the universities are expanding student access to ChatGPT and bringing AI into teaching and learning to prepare graduates for an AI-enabled workforce. The full text could not be read; only the published excerpt is summarized here.

October Layoffs at Three-Year Low; Seasonal Hiring Plans Up 26%

Challenger, Gray & Christmas

U.S. employers announced 43,994 job cuts in October 2026, the lowest October total since 2023 and 71% below the same month a year earlier. Seasonal hiring plans announced through October rose 26%, led by a record announcement from Amazon. In the year-to-date tally, market and economic conditions overtook AI as the leading stated reason for job cuts, after AI had held the top position in earlier monthly reports this year.

Dull, Dirty, Dangerous? Working Conditions and the Limits of Occupational AI Exposure

arXiv (Yu)

Automation is often justified as relief from dull, dirty, and dangerous work. Using 21 O*NET working-condition items across 754 US occupations, this paper tests whether occupations with higher AI exposure actually carry worse conditions. The result runs the other way: higher exposure goes with lower, meaning better, working-condition scores, under both an LLM-based exposure measure and the AIOE index, and with equal occupation weights and employment weights. Adjusting for physical demands, required education, training, experience, and wages shrinks the composite association, and the adjusted confidence intervals include zero.

Learning about Markets from Models: Evidence from an AI Hiring Platform

arXiv (Alidaee et al.)

Paper by Hossein Alidaee, Tarun Khanna and Rembrand Koning on machine-learning models that make decisions, not only inform them. Pooling data across contexts may not raise accuracy when the optimal decision differs by context. A platform keeping distinct models per context captures a model comparison externality: the models jointly show how decision makers respond differently to the same data. Evidence is from an AI hiring platform.

Heimstaden is changing residential real estate management with AI

OpenAI

Customer case study published by OpenAI on its own site. Heimstaden, a Sweden-based residential property company operating across Europe, deployed ChatGPT Enterprise and Codex in property management. The vendor describes the deployment as cutting costs and supporting employees across its European operations. The post is vendor marketing material; the public excerpt gives no figures on headcount, hours or cost savings.

Saving for the Machines

National Bureau of Economic Research

This NBER working paper argues that when machines perform a growing share of tasks, growth depends on how the task structure shapes the return to accumulating machines and how strongly capital accumulation responds. In a standard task model, capital supply sets the size of an automation boom while task structure sets its duration and the wage path. The same technology yields 11.7% annual growth with an optimally chosen saving rate but only 2.2% with a fixed one.

Making AI work for decent work: ILO calls for human-centred approach as Viet Nam accelerates digital transformation

ILO

ILO news from Ha Noi on policy discussions held 26-30 October 2026. It cites ILO analysis from April 2026: 20.8 per cent of employment in Viet Nam, about 11.5 million jobs, is potentially exposed to generative AI, while only about 1.8 per cent is in the highest exposure category; women are over-represented. The ILO names three priorities: lifelong learning, investment in people alongside technology, and strong labour market institutions.

AI Could Do My First Job. Not My Last.

PIIE

Essay by Jason Furman (PIIE), republished from his Econvitals site. Furman argues that most of what the 35-person Council of Economic Advisers did when he chaired it could now be done by about ten people plus AI, and that AI could have done 95 percent of his own junior staffer job in the 1990s. Replaceability depends on task mix, not seniority: junior NEC staff who run meetings are safer than CEA analysts who write on a computer.

What is Work For? Labor, Politics, and Meaning in the Age of AI

arXiv

A cross-national survey of 8,234 working-age adults in the UK, US, Denmark, and India asks what people get from work beyond pay, with the coming AI economic transformation in view. 81% of respondents say they are satisfied with their job, and a majority in all four countries report being strongly motivated by non-monetary components of employment such as purpose, community, and status. On an original "Work Values" index, about a third of workers are intrinsically motivated, and roughly a fifth would keep working even if offered a guaranteed income equal to 120% of their salary; these patterns hold across income levels and across white- and blue-collar jobs. The author concludes that a basic income may be needed to absorb an AI-driven labor shock but would not compensate for the non-monetary benefits lost when people stop working.

Goldman says AI layoffs are starting to leave a ‘long-lasting’ scar, just like the last 40 years of history

Yale Budget Lab

A Yale Budget Lab press entry (October 30, 2026) pointing to news coverage of a Goldman Sachs analysis. The coverage reports that AI-related layoffs are beginning to leave a long-lasting scar on displaced workers, in line with the pattern seen in roughly 40 years of displacement history, where laid-off workers face persistent earnings and re-employment losses. The Budget Lab page carries the headline and date only; the underlying analysis is Goldman Sachs’s.

Credit Saison achieves 954% ROI in a ChatGPT Enterprise pilot

OpenAI

OpenAI customer story on Credit Saison, a Japanese financial services company. According to OpenAI, the company’s ChatGPT Enterprise pilot delivered a 954% return on investment, and the story describes how leaders and employees used the pilot results to drive companywide adoption. Single-company, vendor-published figure; no job or headcount effects are reported in the summary.

Practice Makes Perfect: Tacit Knowledge, Generative AI, and Returns to Schooling and Experience

arXiv

An arXiv paper asks how generative AI changes two prices in the wage structure: the return to schooling and the return to experience. AI makes codified knowledge cheap at the point of production, while judgment built through practice stays with the worker. In the short-run framework, machine-supplied codified knowledge substitutes for the codified part of a worker's human capital and lowers the schooling gradient.

Legora is building the agentic operating system for legal work

OpenAI

An OpenAI customer story on Legora, a legal-tech company building an agentic platform for law firms on OpenAI models, ChatGPT and Codex. Legora uses the models to speed up legal workflows and widen contract analysis, and OpenAI says this shifts lawyers' time toward higher-value work. It is a single-company case written by the model vendor, with no figures in its summary.

The Capability Transfer: Cohort Reversal, Codified Expertise, and the Adaptability Tax of AI Transformation

arXiv

A new arXiv paper shifts the question from which jobs are exposed to AI to what happens to the pipeline that produces experts. It extends the Caosun and Aral (2026) model from a single worker to a firm with junior and senior cohorts. Because junior work is the most codified, automation cuts demand for juniors first; the stock of seniors falls later but further, as the pipeline stops replacing experts who retire.

Retell uses OpenAI models to help power 100M monthly calls

OpenAI

OpenAI customer case study on Retell, a voice-agent company: Retell says its platform handles about 100 million calls a month on OpenAI models, that GPT-6 Astra topped its internal voice-agent benchmark, and that using Codex it reached an annual revenue run rate of $1.6M per employee. All figures are self-reported by one company in a vendor-published post; no independent data on call-center jobs is given.

Training Your Replacement: Worker Incentives and the Pace of Automation

National Bureau of Economic Research

Joshua Gans and Peter Landry model firms that ask workers to train the AI systems that will replace them; the worker can shirk, cooperate with training, or obstruct it while keeping output up. Because finished training ends the job, dismissal loses force as a penalty for shirking, so the firm must raise the worker's rent and the worker gains from being replaced. In the model, AI can create jobs that pay more than they produce and eliminate jobs it never automates; firms paying only wages may slow training, and the authors propose a credible completion bonus equal to the worker's rent without AI as a remedy.

ChatGPT’s expanding world of work

OpenAI

OpenAI published OpenAI Signals data on how people use ChatGPT for work, covering 165 work activities across nearly 100 countries and showing how that usage changes over time.

The Delegation Divide: An Open Index of AI-Agent Delegability across Occupations and Countries

arXiv (Espinal Maya & Jiménez Londoño)

Agent Delegability Index (ADI): two language models rate all 18,796 O*NET 30.2 task statements on a preregistered rubric, mapped to 433 ISCO-08 groups and weighted by ILOSTAT employment in 146 countries. 35.2% of world task time is at least partly delegable to AI agents in 2026; only 2.5% fully. Delegable employment rises with national income through occupational composition. Full delegation peaks for tasks of about 12 minutes; delegability tracks observed AI usage across 101 countries.

How can international labour standards help shape ethical AI at work?

ILO

ILO release for Working Paper 182 (Agustí-Panareda and Samaan) comparing five AI ethics principles (justice, beneficence, non-maleficence, autonomy, explicability) with the ILO normative framework. The two converge on fairness, human dignity, well-being and prevention of harm. Where AI ethics guidelines stress individual autonomy, privacy and transparency, ILO standards add collective rights: freedom of association, collective bargaining and social dialogue. Flagged issues: worker data, algorithmic management, surveillance, AI-assisted hiring decisions.

Uber brings ChatGPT Work to all employees with OpenAI

OpenAI

OpenAI announces that Uber is rolling out ChatGPT Work to more than 30,000 employees, with teams using it to research, analyze data, create presentations, and complete multi-step workflows. This is a vendor-published single-company customer story; it reports a tool deployment rather than any measured effect on headcount, roles, or productivity, and no figures beyond the employee count are given.

How AT&T is rethinking in-house legal work with OpenAI

OpenAI

Vendor case study published by OpenAI on AT&T's in-house legal function. According to OpenAI, AT&T's LegalEdge team is working with OpenAI to make legal knowledge easier to access, redesigning how legal requests are taken in (intake) and how lawyers' judgment is scaled across the company. The excerpt reports no figures on volume, time saved or headcount; this is a single-company account written by the tool vendor, not an independent evaluation.

Middle East, North Africa, Afghanistan and Pakistan Economic Update, October 2026 : From Divide to Opportunity - AI, Jobs, and Growth : Charts

World Bank

This is the charts companion to the World Bank's October 2026 economic update for the Middle East, North Africa, Afghanistan and Pakistan (MENAAP), whose special topic is AI, jobs and growth. The update opens with the conflict that began in the Persian Gulf in February 2026: the effective closure of the Strait of Hormuz cut a route that normally carries about one-fifth of globally traded oil and LNG. Gulf oil output fell from about 26 million barrels per day before the conflict to about 16 million in March.

Tracking AI’s Real-Time Footprint On The Labor Market

Stanford HAI

Stanford HAI's Digital Economy Lab describes its three monthly "AI Economic Indicators" dashboards launched in June 2026: a Transformation Tracker (12 macro indicators), a Canaries Dashboard built with ADP Research, and an Adoption Monitor of generative AI surveys. Employment in AI-exposed occupations such as software development and customer service has fallen more than 4% this year; for workers aged 22-25 it has dropped over 3% a year since ChatGPT's release, most recently 4.4%. None of the 12 macro indicators yet signals explosive growth.

Canaries Dashboard: Employment in AI-exposed occupations shrank in September

ADP Research

September 2026 reading of the ADP Research Canaries Dashboard (Nela Richardson, 2026-10-21). Employment in high AI-exposure occupations fell 0.5% year over year across all ages, while least-exposed occupations grew 1.0%. Workers aged 22-25: overall -2.2%, high-exposure -3.7% (36th consecutive month of year-over-year decline since October 2023), least-exposed +0.2%. Ages 26-30: high-exposure -2.8%, least-exposed +0.9%. Payroll data covering millions of workers in 730+ occupations at tens of thousands of private U.S. employers.

How exposed is work to robotics? Occupational, industry, and geographic measures

arXiv (Sartor, Seamans & Thompson)

Builds the first abilities-based exposure measures for robotics and physical AI, following Felten et al.: 16 robotics capabilities scored by an LLM against 52 O*NET abilities, aggregated for 774 occupations, 250 NAICS industries and US counties. Exposure concentrates in construction, farming, transport and production, and unlike generative-AI exposure it is higher in lower-wage, lower-education jobs. Authors call it an upper bound, not an adoption forecast.

Franchisee reconciles cash in 94% less time with ChatGPT Work

OpenAI

OpenAI customer story about one independent Schlotzsky's and Cinnabon franchisee in the US. OpenAI reports that the operator cut cash reconciliation time by 94% and menu repricing time by 95% with ChatGPT Work, and built a workflow it says could free up $100K in cash flow. The figures come from the vendor, cover a single business, and say nothing about staffing.

When Better Opportunities Backfire: Service Worker Responses to Parallel AI Deployment in Customer Acquisition

arXiv (Fu, Luo, Ramasubbu et al.)

Regression-discontinuity study of an AI voice-agent rollout at a Chinese online health-insurance brokerage, using over 500,000 outreach records. Routing low-intent leads to AI raised the share of high-intent leads on agents' first three calls by about 23 points, yet human customer acquisition on those calls fell about 5 points: agents reallocated early leads 8 points more often and cut those calls about 13% shorter.

Creative destruction: Human-AI competition in an experimental market

arXiv

An experimental study tests two hypotheses for why a human role in production could persist as AI capabilities grow: a durable human advantage in taste and product design, or consumer preference for human involvement in production. Participants acted both as producers, designing computer games with an AI coding agent, and as consumers buying them. Across three studies (N = 1,025), the authors report no evidence for either hypothesis.

Displacement Is Endogenous: AI Workers in a Search-and-Matching Labor Market

arXiv

Simulation of a search-and-matching labor market in which AI systems enter as workers alongside humans: they apply for jobs, pass screenings, get hired and get dismissed, while firms post vacancies and product demand slopes downward. Across 30 seeded runs per configuration, AI entry at 70% of the human wage raises unemployment by 4.1 percentage points (95% CI 3.5 to 4.6) and leaves 12.2 fewer humans employed (CI 10.5 to 13.9). When AI is much cheaper than humans, layoffs account for most of the loss, 7.2 of the jobs lost. These are model results, not observed data.

Tajikistan’s education trade unions address AI and digital transformation to promote decent work

ILO

Tajikistan's education and science workers' union held a two-day seminar in Dushanbe on 13–14 October 2026, with ILO support, on how AI and digitalization affect teachers' work. Sessions covered changing skill needs, heavier workloads, data protection and unequal access to technology. Participants set priorities on digital skills, professional development and social dialogue, and agreed that AI should support teachers, not replace them.

AI Advice Reinforces Traditional Gender Roles

arXiv

Two large-scale audit experiments tested whether four leading large language models give different career-negotiation advice to female and male job candidates. The models recommended lower salary targets for female personas than for male ones and built different negotiation agendas, ranking flexibility, caregiving and well-being higher for women. The paper frames this as a growing issue because professionals increasingly turn to AI for career advice.

Too Late to Prune: Generative AI, the Screen, and Team Creative Decisions

arXiv

Preregistered experiment in which 148 three-person teams were randomly assigned to five schedules of access to the same language model during a product-design task with timed search and selection stages. The authors argue that generative AI moves the bottleneck of creative work from generating options to screening them, so AI's value depends on timing: options that arrive before the team's screen compete for a place in the portfolio, while options that arrive after it are kept without being screened.

Sophos cuts threat investigation time by 96% with OpenAI Daybreak

OpenAI

OpenAI customer case study (company-published): OpenAI says security firm Sophos uses OpenAI's Daybreak to cut cyber-threat investigation time by 96% and to automate 52% of its managed detection and response (MDR) cases while preserving human oversight. Both figures are the vendor's own claims about a customer; this summary rests on the feed excerpt only, so the baseline, measurement period and method behind them are not stated here.

AI training helps turn technical skills into enterprise opportunities

ILO

ILO article on its AI4MSME training in the Philippines. Vocational trainees at TESDA Region 2 in northern Luzon (Cagayan, Isabela and Nueva Vizcaya) are taught to use AI tools for menu development, costing, marketing and customer proposals, with the aim of linking the technical skills they already have to opportunities to start an enterprise.

How Oracle turns days of work into minutes with ChatGPT and Codex

OpenAI

OpenAI customer case study (company-published): OpenAI says Oracle uses ChatGPT Work and Codex across recruiting, engineering and operations to turn specialist knowledge into fast, repeatable workflows, and the headline claims work that took days now takes minutes. Based on the feed excerpt; the page itself was not read.

Understanding Interfirm AI Talent Flow Networks through Online Professional Profiles

arXiv

This arXiv preprint reconstructs the inter-firm network of AI talent flows for 2010-2022 from about 535 million employment records across 58 countries. AI talent inflows are increasingly concentrated among a small set of leading firms, with stronger preferential attachment than in general labor mobility, yet central positions change hands far more often than in non-AI networks. AI network centrality is associated with higher enterprise value after accounting for workforce scale, assets and fixed effects. Summary based on the abstract; the paper was not read.

Want Employees to Embrace AI? Stop Selling It as a Productivity Tool

Stanford HAI

Stanford HAI reports on research by Arvind Karunakaran (Stanford), Roshni Raveendhran (University of Virginia) and Tami Kim (Dartmouth) on how a manager's framing of AI changes employee uptake. The team interviewed 184 employees at three companies that had adopted generative AI tools, then spent nearly two years observing two near-identical divisions of a law firm. In the division where the manager asked paralegals which tasks they found boring, instead of pitching faster drafting, paralegals used the tool "LawBot" 58% more and experimented with it 70% more. Based on the HAI article; the underlying paper was not read.

South Asia Economic Update, October 2026 : Adopting AI for Growth

World Bank

World Bank, South Asia Economic Update for October 2026, with "Adopting AI for Growth" as its theme. The report projects South Asia to remain the fastest-growing emerging and developing region, despite costly energy imports and limited participation in global value chains for AI-related goods. Named downside risks are a severe El Niño, persistently high energy prices and global financial strains. It also notes that the region is ageing unusually fast with little financial security for older people, which makes creating more and better jobs a priority.

Managing the AI Transition: Upskilling and Reskilling Asia’s Workforce in the Age of Automation

Asian Development Bank (ADB)

This Asian Development Bank Institute (ADBI) publication addresses upskilling and reskilling of Asia's workforce as artificial intelligence is rapidly integrated into the economies of Asia and the Pacific and transforms their labor markets. It warns that without policy intervention, developing countries risk deindustrialization and the loss of demographic dividends. Summary based on the two-sentence publication feed abstract; the publication itself was not read.

How Heavily Are PhD Students Offloading Their Scientific Writing to AI? Evidence from US STEM PhD Dissertations (NBER WP 35859)

NBER

Gross, Shvadron and Zhang measure AI-generated writing in the dissertations of a near-census of US STEM PhD graduates from 2019 to 2026. AI writing is absent before 2023 and present in 29% of dissertations filed in 2026, more often among students from non-English-speaking countries and at lower-ranked programs. Within the same program and cohort, graduates with AI-written dissertations are less likely to enter academic research jobs and more likely to take non-tenure-track or industry jobs. Based on the NBER abstract page; the full paper was not read.

Right Order, Wrong Scale: Auditing LLM Judges for Occupational AI Measurement

arXiv

This arXiv preprint audits LLM judges used to assess whether AI outputs meet workplace requirements, arguing that agreement on response rankings does not establish agreement on acceptance rates or occupational aggregates. It introduces O*NET-BENCH, built from an existing survey of 45,796 worker ratings, and evaluates 33 pre-existing judge configurations across six model families on 4,501 test ratings; 25 reach tie-aware pair accuracy of at least 0.60. Based on the truncated feed abstract; the paper was not read.

The eternal complement

OpenAI

OpenAI essay arguing that advanced AI may matter most for the routine work behind breakthrough ideas, and that execution could shape the next economy and the pace of progress.

How Albertsons Companies is reimagining retail from the inside out

OpenAI

OpenAI customer case study (company-published): grocery retailer Albertsons Cos. is using ChatGPT Enterprise and the OpenAI API to help teams work faster and make grocery shopping easier for millions of customers.

Job Cuts Fall in September; Hiring Plans Up 3% Over 2025 On Weak Early Seasonal Hiring

Challenger, Gray & Christmas

Challenger, Gray & Christmas's monthly count of announced U.S. layoffs: employers announced plans to cut 43,281 jobs in September 2026, 18% fewer than in August. On the hiring side, weak early seasonal hiring plans held year-to-date hiring announcements to just 3% above the same period of 2025. The figures are announced plans tallied by an outplacement firm, not realized job losses, and they move with the timing of individual corporate announcements.

The Den frees up 10-15 hours a week to grow with ChatGPT Work

OpenAI

OpenAI customer case study (company-published): social club The Den says ChatGPT Work frees up 10-15 hours a week, with grant applications prepared in 2 hours instead of 3 days and liquor-license materials in 3 hours instead of 4 days as it opens a new location.

Maintaining Human Verification Capacity under Automation

arXiv

A theoretical paper on a quiet risk of automation: human verification relies on expertise that has to be kept up before it is ever needed. The model links three things — reliance on automated checks, investment in human checking ability, and performance when automation is interrupted. Its central result is that better automated checking shrinks the error reduction gained from an extra unit of human skill, which can weaken the incentive to preserve independent expertise even when it lowers the best achievable expected cost. The paper is a formal model and reports no empirical estimates.

Helping small businesses put AI to work

OpenAI

OpenAI announced a partnership with America's SBDC (Small Business Development Centers) to expand hands-on AI training and local support for small businesses, released alongside a company report on how small teams are using AI. This is a corporate announcement rather than independent research.

Can we predict the jobs robots will do?

Anthropic

Anthropic economists build a robot exposure index from O*NET tasks rated by how structured an environment a robot needs. Robots can already do 74% of US physical tasks (34% of all working hours), but are cheaper than human labour for only 0.3% of work; at past 3%/yr price declines, reaching 10% would take about 40 years. Driving and packing jobs are most exposed, nursing and repair least.

When Less Is More: Managing AI Adoption with Adaptive Incentive Design

arXiv (Gong et al.)

A large medical-device firm required about 5,000 employees to send at least 200 queries a month to an internal AI assistant. Use rose, but so did gaming: queries bunched at the threshold and 31 percent were repeated or off-task. Lowering the target to 100 cut query volume by 30 percent, about 90 percent of it from repeated or off-task queries, with no significant change in genuine work queries, while sales employees' monthly sales rose 7 percent.

Economic Governance of Autonomous Agents and Robots: Factor-Origin Accounting and Social Automation Funds

arXiv (Syed et al.)

Argues that tax systems built on wages and payroll erode as autonomous agents, foundation models and robots decouple output from human work hours. The paper proposes CivicDividendOS, a fiscal framework that attributes value across labor, AI, robotics, data and capital, levies a Social Automation Contribution on automated value added with offset credits for upskilling, and recycles revenue into payroll relief and a public wealth fund. It is a conceptual design illustrated in urban logistics, not an empirical estimate.

AI-based matching improves refugee employment in a double-blind randomized trial

arXiv (Bansak, Hainmueller, Hangartner et al.)

In a double-blind trial run by the Swiss State Secretariat for Migration from 2020 to 2023, about 2,000 refugee cases were randomly given either an algorithmically employment-optimized canton recommendation or one approximating existing procedures, with officers keeping final say. The share of months employed over three years rose 2.2 percentage points (3.9 pp for 2022-2023 cohorts), and the 36-month employment rate rose 5.2 pp, comparable to hundreds of hours of intensive language training.

The Macroeconomic Effect of AI: Sizing the Software Engineering Channel (NBER WP 35793)

National Bureau of Economic Research

Blumenfeld, Hazell, Lian and Schaab measure how AI affects the economy through one channel: software engineering productivity. They build a forward-looking, real-time measure from financial markets, estimating each firm's stock-return sensitivity to an AI stock market index and how it varies with the share of firm payroll in software engineering, then use a model to map that relationship into productivity gains. The period studied is November 2022 to December 2025.

The Early Impacts of AI on Employment among Recent College Graduates (NBER WP 35796)

National Bureau of Economic Research

Fairlie and Wu use CPS microdata to produce what they describe as the first estimates of AI's effect on unemployment among recent college graduates in June, July and August 2026. The abstract says unemployment rates run especially high in summer months and that 2026 might be the first year with workplace AI use widespread enough to detect effects on this group. It notes there is no consensus on magnitude or timing.

From exposure to opportunity: Why skills shape the employment effects of new technologies

ILO

Using ILO harmonized microdata for more than 1,000 subnational regions in 69 countries, the analysis estimates how potential exposure to 40 emerging digital technologies, including industrial automation, robotics and machine learning, affects employment rates. On average exposure is associated with employment gains, but unevenly: concentrated among prime-age workers, reaching low-educated workers in low-income countries and high-educated ones in high-income countries. Gains are larger where local labour demand embeds stronger advanced cognitive, basic computer, finance, writing, project and people-management skills, even after controlling for education (Delaporte and Escudero, forthcoming).

Old workers, young machines: can AI and automation offset population ageing?

Bank for International Settlements

BIS Bulletin No 135 (Aldasoro, Doerr, Rees) asks whether AI and robots can make up for shrinking, older workforces. The catch is where ageing happens: automation fits best in industries with young staff such as finance, while agriculture and health, which are older and employ many people, offer less room. A scorecard covering more than 130 economies flags several of the fastest-ageing countries as the ones least able to lean on automation.

Grab and OpenAI bring practical AI skills to Southeast Asia

OpenAI

OpenAI and Grab launched "GO Forward with AI", a regional programme to help 30,000 Grab partners across Southeast Asia build practical AI skills.

Canaries Dashboard: Employment in AI-exposed occupations slowed in August

ADP Research

ADP's monthly Canaries Dashboard reports that employment in AI-exposed occupations slowed in August 2026, with early-career workers continuing to lose ground.

AI plays a role in weak labor market for college graduates

Federal Reserve Bank of Dallas

Dallas Fed researchers Samuel Dodini and Tucker Smith link Texas administrative records, Lightcast job postings, O*NET and Anthropic task data to measure how generative AI is reshaping outcomes for new college graduates. Graduates from AI-exposed majors were 1.7 percentage points less likely to find a Texas job within a year and earned 5% lower wages than peers in less-exposed fields. They were also 1.4 points more likely to enter graduate school, and undergraduate enrollment in exposed fields fell 4.8% between fall 2024 and fall 2025. The authors conclude that generative AI has reduced demand for roles whose tasks it completes more efficiently than humans.

AI adoption in Chinese enterprises boosts productivity but raises concerns about jobs and skills

International Labour Organization

ILO news summary of its 15 September 2026 brief on AI adoption in Chinese enterprises (21 firm interviews, 1,591 professionals surveyed). Firms mostly add AI to hybrid workflows where people still guide its use. Productivity gains are self-reported, not independently verified: one insurer's 300 customer-service staff went from 6,000 to 15,000 issues a day; a smart factory reported 30% higher efficiency. 56% of professionals see adoption as inevitable, 39% expect lower income. Routine clerical, administrative and customer-service roles face the most displacement pressure.

Parallel cut research time and cost in half with GPT‑6 Astra

OpenAI

OpenAI customer note: GPT‑6 Astra allowed Parallel's agents to research and synthesize labor-market data in half the time and at half the cost compared with prior models. Summary is based on the feed excerpt only; the full article was not accessible at registration.

Expanding OpenAI Academy with new learning paths

OpenAI

An announcement published by OpenAI. According to OpenAI, OpenAI Academy is adding new learning paths for five audiences (employees, developers, leaders, educators and students) so that people can build practical AI skills and demonstrate them. This is the company's description of its own training offering; this entry summarizes the announcement only and does not assess the curricula or any outcomes.

How V7 gives AI agents institutional memory

OpenAI

A customer story published by OpenAI. According to OpenAI, V7 uses GPT-5.6 to turn a company's scattered files into context that AI agents can use, so the agents can complete complex work whose outputs are linked back to the underlying sources. OpenAI frames this as giving agents "institutional memory". The claims are the vendor's own and are summarized here from its post.

How Cooley is accelerating IPO work with ChatGPT

OpenAI

A customer story published by OpenAI. According to OpenAI, the law firm Cooley built "GO Public" with ChatGPT Work to bring intelligence to the IPO process, helping lawyers surface issues earlier and focus their judgment where it matters most.

Better Technology, Worse Motivation: Generative Artificial Intelligence’s Mediocrity Trap

Asian Development Bank (ADB)

This ADB paper examines how generative artificial intelligence (AI) affects both productivity and motivation in creative work. Summary based on the publication feed abstract; the paper itself was not read.

Hex turns complex analysis into visual reports with GPT‑6 Astra

OpenAI

A customer story published by OpenAI. According to OpenAI, GPT-6 Astra helps Hex's data agents turn answers into interactive visualizations that employees are proud to share.

Distributional Intersectional Fairness in AI-Supported Job Matching

IAB (Institute for Employment Research)

An IAB paper analyses how machine learning methods change the distribution of outcomes between intersecting social groups, in the setting of AI-supported job matching. Summary based on the institute's one-sentence announcement; the paper itself was not read.

Workforce policy for the age of AI: Recommendations from the economic literature

Brookings Institution

Ben-Ishai and Thompson (MIT FutureTech), Brookings Economic Studies, 2026-09-15. Four findings from the economic literature: exposure is not commercial automation (models complete 3-4 hour tasks at 65% success as of end-2025; 10x task duration cuts performance 11%; firms would automate only 23% of automatable computer-vision tasks at current costs); the question is how AI changes the value of expertise (accountant vs family physician example); adoption requires knowing when to trust AI; reliability gains (failure halving ~2.5 years) will widen autonomous use. Five policy directions: prioritize displaced workers (displacement costs 2.8 years of prior earnings; low-wage workers -13% after six years) and high-productivity opportunities, domain-specific employer-partnered training, align programs with expertise shifts, expand apprenticeships (pay-for-performance, LEAP Act), federal wage insurance (TAA precedent). Authors concede training evidence is weak. Absolute mobility 90% (1940s cohort) to 50%.

Artificial intelligence adoption in Chinese enterprises: Productivity effects, workforce implications, and policy challenges

International Labour Organization

ILO Research Brief (Ernst with Renmin University co-authors, 2026-09-15, CC BY 4.0). 21 firm interviews + survey of 1,591 professionals. Reported gains: recruitment 30->13 days, 300-agent CS floor 6,000->15,000 issues/day, spec-reading team 10+->2-3, lighthouse factory +30%. 13 of 21 firms report no quantified gain. Survey: 56% inevitable, 47% net job creation, 39% expect income decline. Self-reported, purposive sample, no counterfactual.

Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews

arXiv (Jabarian & Henkel)

Preprint (arXiv 2607.28222v2, 2026-09-11) by Jabarian (CMU) and Henkel (Erasmus). Natural field experiment at PSG Global Solutions (Teleperformance subsidiary) in the Philippines: 70,884 entry-level customer-service applicants randomly assigned to human-recruiter or AI voice-agent interviews; humans made all hiring decisions. Offer rate 8.70% (human) vs 9.73% (AI), +12% relative; job starts +18%; retention +18-19% at 2-4 months; no productivity difference on three firm metrics. Mechanism: 'controlled variance' (guideline similarity 0.59 vs 0.43). 78% chose AI when given the choice, with negative sorting. Lead author became unpaid Chief Economist at the partner firm after data collection.

Traditional labor market data isn't keeping up with jobs in the 'frontier economy' (Lipson & Schwartz)

Brookings Metro

Five blind spots in NAICS/SOC/O*NET data for frontier jobs: supply chain nearly triples semiconductor postings; biopharma machine-maintenance workers 26K (2008) to 57K (2025) with no SOC code; nuclear technician postings +68% in 2025 vs BLS -8% projection later revised to +1%. Revelio Labs data.

Record of proceedings: Technical meeting on challenges and opportunities for promoting decent work, productivity and a just transition arising from AI in the manufacturing industry (TMDWAI/2026/9)

International Labour Organization

Record of the 13-17 April 2026 ILO tripartite technical meeting (115 participants, 54 countries) that unanimously adopted the first ILO conclusions on AI in manufacturing (TMDWAI/2026/8, to Governing Body Nov 2026). 11 of 17 paragraphs amended: 'when applicable' inserted into worker-consultation clause, guidelines recommendation dropped (employers + US), risk assessments 'encouraged', surveillance/algorithmic 0 mentions. Background report: 489-494M manufacturing jobs, 16.5% GDP; employer barriers cost 58% / skills 43% / regulation 19%.

The Challenger Report August 2026: Job Cuts Up 58%; Consumer Products, Food Lead

Challenger, Gray & Christmas

August 2026 US job cuts totaled 52,881 (+58% MoM, -38% YoY; quietest August since 2022). AI fell to the 4th-most cited reason (3,462 cuts, lowest since Dec 2025) after leading March-July; AI remains #1 YTD at 116,175 (~22% of 529,914 total cuts). Hiring plans rose 725% YoY to 12,325; YTD 119,825, strongest Jan-Aug since 2023.

Job postings show early signs of AI automation impact

Federal Reserve Bank of Dallas

Dallas Fed Economics article (Dodini & Smith, 2026-09-01) linking Lightcast job postings to Anthropic's task-based AI exposure index. Postings for more-exposed occupations fell ~5% by end-2023 and ~7-8% by 2025-26 relative to less-exposed occupations within the same industry (per 10pp exposure difference). Incumbent Texas firms cut postings 8-9% by early 2026 and shifted composition away from automatable tasks (-2pp, ~50% of mean). Aggregate Texas postings reduced ~1.8% (2024) and 2.6% (2025). Two-thirds of Texas firms use AI (TBOS May 2026, vs 40% two years earlier).

Does AI Assistance Enhance or Erode Expertise? Evidence from a Three-Month Field Experiment in Patent Drafting (NBER WP 35720)

National Bureau of Economic Research

Pre-registered 3-month RCT (Autor + 6 Google co-authors): 133 patent lawyers at 11 US IP firms, 2:1 AI (Google InFlow) vs control, blinded expert scoring. AI raised drafting quality 0.34 SD (10d) / 0.38 SD (90d), juniors most; 10-min time saving on 112-min task. Unassisted redlining at 90 days: treated +0.32 SD, entirely among seniors (+0.45 SD); juniors ~0 on average with bifurcated distribution. 91/133 completed (32% attrition), 15/91 flagged possible AI use on no-AI task, results robust.

인공지능(AI)는 일자리 파괴자 아닌 직무 재구성의 파트너… '인간 중심 인공지능(AI) 발전전략' 세워야

Ministry of Employment and Labor

Press release from Korea's Ministry of Employment and Labor announcing the Summer 2026 issue of "Employment Issues" (고용이슈), published by the Korea Employment Information Service. A survey of 2,297 establishments puts AI adoption at 28.6%, which the release calls an early stage; among adopting firms, 75.6% use generative AI. Among workers under 30, the 2025 employment index (2022 = 100) stood at 86.6 for the high-AI-exposure group against 92.5 for the zero-exposure group, a gap of 5.9 points, though the authors note the high-exposure decline began before ChatGPT. A separate survey of 305 incumbent workers reports an average self-rated skill gain from AI of 2.97 out of 5, highest for efficiency (3.40) and lowest for collaboration (2.37). The issue frames AI as a force that reshapes tasks rather than eliminating jobs and calls for a human-centered AI development strategy.

AI Transformation Requires Redesigning Work, Not Cutting Roles

Harvard Business Review

A Harvard Business Review article arguing that many companies are making AI-driven workforce cuts faster than the evidence warrants. It says early AI-driven layoffs have often failed to deliver expected returns, exposed gaps in organizational knowledge and created new demands for human oversight, and that some employers have reversed course. The proposed alternative starts from the work rather than headcount targets: break roles into tasks, identify where AI genuinely improves performance, and redesign processes around the right mix of human and technological capabilities.

The Pulse Beneath the Job Title: Monthly Readings of Requirements and Tasks from 750 Million Chinese Job Ads

arXiv (econ.GN)

Preprint analyzing 752.6M Chinese job ads (2022-2026, 5 platforms) into catalogs of 20,721 requirements and 44,479 tasks with monthly LLM-exposure readings. Key finding: AI-exposed occupations lose posting share (slope -1.33) far faster than their task categories decline (-0.83) — adjustment runs through which jobs are posted, not job content. Within-title exposure spreads up to 17%-48% by wage band.

Sophistication in GenAI Use: Field Evidence from a Large Firm

arXiv

Field study of 713,564 prompts from 3,925 KPMG back-office employees over 8 months of 2025. Sophistication rises with seniority and peaks in Strategy/Digital Innovation/PM; Accounting & Finance lowest. No improvement over time; formal training effects vanish after the training month.

Employment Projections: 2025-2035 (with new AI exposure categories data product)

U.S. Bureau of Labor Statistics

BLS projects +5.9M jobs 2025-35 (170.3M to 176.2M, +3.5% vs +10.9% in 2015-25) and debuts a four-category AI exposure classification (Low/Moderate/High/Very high) for 831 occupations, combining 3 theoretical measures (Felten et al., Eloundou et al., Eisfeldt et al.) with 2 observed telemetry sources (Anthropic Claude, Microsoft Copilot). Office/admin support -4.0% (-752,100, largest group decline); healthcare 37.0% of all new jobs; nurse practitioners +41.0% fastest.

Monocultural Biases: Correlated biases in large language models lead to unequal systemic exclusion rates in hiring

arXiv

Bone, Stephany and del Rio-Chanona test ten LLMs used to screen job applicants, comparing base and post-trained versions. Post-trained models were 3.6% less likely to call back older applicants (in 8 of 10 models) and made far more correlated decisions. That consensus raised systemic exclusion rates from 5.6% to 17.3%, with intersectional rates of 12.2-21.7%, driven mainly by age-based bias.

Normative boundaries of AI in scientific work: Evidence from PhD researchers

arXiv (Angelini & Lyrvall)

Latent class analysis of 3,785 PhD students (Nature Graduate Survey 2025) finds 4 attitudinal profiles: division-of-labour 44%, status-quo 34%, all-purpose 16%, undecided 7%. AI accepted for literature tasks (60%/48%) but resisted for writing, data analysis and experiment design (33%/33%/36%). Weekly/daily AI use strongly predicts leaving the status-quo profile (-1.911, p<0.01).

AI adoption and the productivity promise: what workers report

European Central Bank

ECB Blog (Dias da Silva, Lebastard, Sondermann, 2026-08-26), cited in Lagarde's 2026-09-14 speech. ECB Consumer Expectations Survey (~20,000 respondents, 11 euro area countries): workers using AI at work 26% (2024), 41% (2025), 52% (2026), ~3 days/week. Highly educated 61% vs lower education 37%; younger +20pp. Median user saves 3 hours/week (7.7% of working time); only 48.8% use and save time, so economy-wide gain ~3.8%. Coding saves ~8 hrs/week but used by ~8%. Positive sentiment fell 43% to 41%. 41% of workers uninterested (34% of managers); ~half of firms plan AI training (SAFE).

What Work Does Generative AI Do? (Bick, Blandin, Deming, Schumacher)

National Bureau of Economic Research

First task-level genAI adoption indexes from a nationally representative US survey (RPS, 4 waves Aug 2025-May 2026). Adoption is widespread but shallow: >80% of occupations and >40% of tasks exceed 20% adoption, yet fewer than half of workers adopt within most tasks. Exposure scores explain 5-53% of occupation variation; chat-log measures over-classify generic tasks (correlations with survey 0.10-0.34).

Labour demand volumes by Standard Occupation Classification (SOC 2020), UK: January 2017 to July 2026

Office for National Statistics

ONS online job advert volumes by 4-digit SOC 2020 (Textkernel-collected, ONS de-duplicated). Total UK postings peaked at 13,780,433 in 2022 and fell to 8,312,308 in 2024 (-39.7%), 27.3% below 2019. Steepest declines cluster in graduate-entry digital/consulting roles; early education and childcare grew. Official statistics in development; Mar 2026 suppressed, several months partially imputed.

Canaries Dashboard: Employment in AI-exposed occupations rose slightly in July

ADP Research

July 2026 reading of the ADP Research / Stanford Digital Economy Lab Canaries Dashboard (Nela Richardson, 2026-08-19). Employment in high AI-exposure occupations rose 0.1% year over year, the first positive all-ages reading in the monthly series, while least-exposed occupations grew 1.1%. Workers aged 22-25: overall -1.3%, high-exposure -3% (34th consecutive monthly decline since October 2023), least-exposed flat. Ages 26-30: high-exposure -2.1%, least-exposed +1.8%. Payroll data covering 730+ occupations at tens of thousands of private U.S. employers.

CentaurBench: Benchmarking LLM Capabilities on Augmenting vs. Automating Real-World Work Tasks

arXiv (UC Berkeley Haas, DIAL)

CentaurBench (arXiv:2608.18554v1, UC Berkeley Haas DIAL) separately benchmarks LLM automation vs augmentation across seven economically grounded tasks. Automation and augmentation rankings correlate only rho=0.48 (p=0.187, n=9 models); the augmentation winner differs from the automation winner in 5 of 7 tasks. The unaided GPT-3.5-Turbo worker outranked every assisted condition on operations research, tax preparation and travel planning. Only GPT-5-Mini beat the no-guidance baseline on average (mean rank 3.66 vs 3.79). Inter-judge agreement 71.0% across 6,265 comparisons (74.5% automation, 67.8% augmentation).

[제2026-19호] 청년고용 위축, AI 탓인가? 변화하는 경력 사다리와 대응 과제

한국은행 (Bank of Korea)

BOK Issue Note 2026-19 (Oh Sam-il, Oh Young-sik): Korean youth (15-29) jobs fell 285,000 from June 2022 to June 2026, 268,000 (94.0%) in AI-high-exposure industries, while workers in their 50s gained 230,000 (75.2% in the same industries). Youth employment -31.4% information services, -27.4% publishing, -16.6% computer programming, -11.6% professional services. Automation-heavy industries show steeper youth declines; augmentation-heavy do not. Graduate youth unemployment 7.0% vs 5.4% since Nov 2022. Monthly youth outflow from high-exposure industries +32% (3.7k to 4.9k). Authors frame AI as an accelerant on top of experienced-hire preference, pandemic over-hiring and remote work, not the sole cause; prescribe career-ladder redesign.

Stranded credentials: how a skill-signaling market absorbed generative AI

arXiv (Song Yao)

Audit of 444,698 Kaggle participations (2010-2026) across upload- and code-competition formats. Medals predict subsequent leaderboard performance almost entirely in the first year after being earned. Upload-competition medal stocks lost 82% of informativeness, but institutional stranding (format exit predating AI) explains half to three quarters of that loss. Official lifetime-count credential tiers discard 13-16% of medal information; a recency-weighted index built on pre-AI data alone outperforms them. Preprint, not peer reviewed.

Will AI take Filipino jobs? The answer depends on what we do now

International Labour Organization

Op-ed by Khalid Hassan, Director of the ILO Country Office for the Philippines. Entry point to the ILO February 2026 policy brief "Generative AI and Jobs in the Philippines" (Phu Huynh, PSA LFS Q1-2024 microdata): 27.7% / 12.7m of Philippine employment exposed to GenAI (highest in ASEAN), only 3.6% (1.7m) in Gradient 4, of which ~78% are clerical codes. IT-BPM employed 1.8m in 2024 (1.6m contact centres). ~2/3 of IBPAP member firms use AI vs 14.9% of Philippine firms overall. Fact-check: the 6.1% youth figure quoted in the op-ed is the World row of GET for Youth 2026 Table 2.2, not a Philippine figure; the Philippine youth Gradient-4 rate is 4.2% (217,200 jobs).

Changing landscape of skills in the age of AI

ILO / IAG on TVET (ETF, Cedefop, Eurofound, EC, ILO, UNESCO)

Joint six-agency (IAG TVET) synthesis on how AI adoption shifts the variety and depth of skills demand: rising higher-order cognitive and socio-emotional skills, small AI workforce (0.3-5% of employment), AI literacy as a new foundational skill, and training supply lagging demand.

Developing a strong workforce requires collaboration

Federal Reserve Bank of Dallas

Dallas Fed SVP Roberto Coronado on AI impact for jobs not requiring bachelor degrees: administrative and IT roles greatly affected, truck drivers and auto mechanics less immediate change; short-term credentials add about $5,000/yr for high school or associate holders vs $2,600 for bachelor holders (Census-controlled).

An evidence review of worker retraining (Anthropic Institute Working Paper 2026-01)

The Anthropic Institute (arXiv 2609.07011)

Roodman & Massenkoff (Anthropic Institute WP 2026-01, arXiv 2609.07011) meta-analyze 146 impact estimates from 56 US randomized job-training trials since 1973: employment +1.7pp (years 3-5), pre-tax earnings +$791/yr per person offered, cost $13,598/participant, narrow B/C 1.44, government recoups ~3/4 of outlay. JTPA/WIA/Job Corps small or null. Sector programs (Project QUEST, Per Scholas, Year Up): earnings +$3,000-4,000/yr family average, $5-10K at top programs, B/C 7.18, but accept ~20% (Per Scholas 7.1%) of applicants and have proven hard to replicate.

AI and the research assistant: Evidence from economics working papers

arXiv (Fourie)

Working paper by Johan Fourie using 24 economics working-paper series through October 2026. It compares papers by the same economist that do and do not disclose AI use, with a detector of AI-like prose, and counts named research assistants in acknowledgements, whose data cleaning, coding and replication work language models now perform. Disclosing papers credit 0.548 more named assistants. v2 posted 2026-11-02 (v1 2026-08-09).

World Development Report 2026: The Promise of Artificial Intelligence

World Bank

The World Bank's annual flagship report assesses AI's potential for developing economies through an adopt/adapt/advance framework. Key findings: 14.2% of high-income-country jobs at generative-AI automation risk vs 4.5% in low- and middle-income countries, while productivity-gain potential is nearly equal (18.7% vs 16.2%); middle-income countries reached 50% of ChatGPT global traffic within six months of launch; "small AI" field evidence includes +40% daily diabetic eye screenings in Bangladesh, ~1 year of additional learning at ~US$5/student via SMS math tutoring in Ghana, and up to US$560 farmer savings from AI monsoon forecasts in Telangana.

Occupational Convergence or Divergence? Mapping Labor Market Structural Shifts Driven by AI Penetration

arXiv (Hilman & Koltai)

Analysis of LinkedIn/Indeed/Glassdoor postings (Dec 2025-May 2026, 10 countries) finds 75-85% of AI-related vacancies concentrated in STEM occupations; health/education, services, production/logistics, and public service each below 5%. AI skill demand concentrates at entry level, acting as a precondition for occupational access. Bifurcated pattern: convergence within the STEM core, divergence from the rest — consistent across Global North and South.

Use of generative artificial intelligence tools among Canadian workers, March 2026

Statistics Canada

First results from a new recurring Labour Force Survey supplement on AI use at work (reference period March 2026, workers aged 15-69 in the ten provinces). 35.9% of workers used generative AI tools for work in the previous 12 months; 41.6% used at least one AI or automation technology. Adoption sorted by an exposure-by-complementarity taxonomy: high exposure/high complementarity 53.8%, high exposure/low complementarity 45.9%, low exposure 14.2%. Industry ranged from professional, scientific and technical services 65.6% to accommodation and food services 16.3%. Within high-complementarity occupations, workers aged 15-24 used AI least (39.1%) versus 56.9% for ages 25-54. Depth of use is shallow: among users, 63.5% some but not most tasks, 24.9% almost no tasks, 8.0% most, 3.6% almost all. Among the 64.1% of non-users, 56.0% said it was not applicable to their work and only 5.8% cited a skills gap.

Unbundling jobs: Measuring the value of tasks in an AI economy

ADP Research

ADP Research with Stanford Digital Economy Lab (Richardson, Wang) linked 5M+ ADP client job postings to 9M+ workers in payroll data; analytic sample ~7,000 workers in selected IT jobs 2019-2025, 20,000+ worker-year observations across 600+ employers. Postings mapped to 25 O*NET Intermediate Work Activities; wages regressed on task indicators controlling for age, gender, year and firm effects. 8 activities associated with higher wages (incl. advising others on design/use of technologies, designing databases), 6 with lower wages (incl. diagnosing system or equipment problems, monitoring operation of computer or information technologies). 5 tasks showed reduced compensation in 2023-2025 vs 2019-2022, including "develop models of systems, processes, or products" which otherwise ranks high-value. No effect sizes, standard errors or significance tests published. Conclusion: AI's biggest and most immediate effect is on tasks, not entire occupations.

How Houston finds workers in an AI boom

Federal Reserve Bank of Dallas

Dallas Fed President Logan Houston visit: AI chatbot pilot matching job seekers to employers by skillset; GE Vernova Houston Learning Center turbine training, high schooler to field engineer in 7-8 years; Houston payrolls +2.1% annualized over 3 months ending May 2026.

AI and the global economy: implications for central banks (BIS Bulletin No 130)

Bank for International Settlements

BIS Bulletin 130 finds the AI boom is driving a large, increasingly debt-financed investment surge (~1% of GDP in the most exposed economies), with an uncertain and uneven productivity payoff. Labor displacement is limited so far, with early substitution signals in call centres and business centres; widespread wait-and-see adoption is consistent with a low hiring, low firing dynamic. AI blurs cyclical signals, complicating monetary policy.

Who Takes the Hit? The Uneven Impacts of Generative AI on Labor Demand Across Countries

World Bank

WDR 2026 background paper analyzing 555 million Lightcast job postings across 84 countries (2021Q1-2025Q2). In high-income countries, postings for above-median GenAI-vulnerability occupations fell 5.8% relative to less substitutable occupations after ChatGPT; in middle- and low-income countries the impact is small and statistically insignificant. Displacement is larger with greater GenAI adoption, schooling, English proficiency, and digital services trade specialization.

Canaries Dashboard: Employment in AI-exposed occupations contracted in June

ADP Research

June 2026 reading of the ADP/Stanford Digital Economy Lab Canaries Dashboard. Employment in high AI-exposure occupations fell 0.2% year over year while least-exposed occupations grew 0.6%. For workers aged 22-25 high-exposure employment fell 4.3%, a 33rd consecutive monthly decline since October 2023; least-exposed employment for that age band was flat. For ages 26-30 the figures were -2.6% and +0.9%. Based on payroll files of tens of thousands of private US employers, ~4 million workers, 730+ occupations.

Economic Futures Research Fund: Research Agenda

Anthropic

Anthropic published the research agenda for its $200M Economic Futures Research Fund: grants of $5-30M (minimum $1M) to universities, research institutes, and nonprofits for large-scale RCTs across five areas — firm-level AI impact, transition support, income support modernization, pre-displacement worker equity, and public investment evidence.

Artificial intelligence in UK businesses: 2023 to 2026

Office for National Statistics

ONS official statistics in development. UK business AI adoption rose from ~12% (Sept 2023) to ~35% (June 2026) among 10+ employee businesses, but average AI technologies per adopter moved only 1.4 to 1.6 and just 10% describe use as extensive. Role impacts concentrate in creative/design and administrative/clerical work; ~7% of medium-sized businesses report headcount decrease. BICS Wave 159, 5-28 June 2026, 26.7% response rate.

The Last Ten Per Cent: How task-level AI exposure fits into a jobs forecast

Economic Innovation Group

Joshua Gans (Rotman/U. Toronto) reframes the "AI automates 90% of a job" claim as an economic question: whether the remaining 10% expands (focus effects, expertise recalibration, org redesign, demand elasticity) or shrinks headcount is indeterminate. Argues task-level exposure matters more than job-title automation, and that exposure/applicability scores are starting points, not employment forecasts — converting them requires investigating expertise thresholds, job bundling, market demand, and work intensity that payroll data obscures.

Old Skills for new technologies?

International Labour Organization

ILO blog by economists Isaure Delaporte and Hannah Liepmann (Research and Statistics Department) arguing that foundational cognitive and socio-emotional skills, more than specialist AI expertise, decide who reaches quality jobs as AI spreads. Citing the ILO's online-vacancy analysis, they report that AI and machine learning skills make up only around 1% of skills requested in Brazil, Egypt, Jordan and the United Arab Emirates, and less in Morocco, South Africa and Uruguay. In the UAE, vacancies asking for AI skills carry a 17.6% wage premium, yet 84% of them also ask for core cognitive skills and 88% for social skills. The authors call for lifelong learning for experienced as well as entry-level workers, and for better pay recognition of socio-emotional skills in care work.

The Impact of Artificial Intelligence on Israel's Labor Market (Selected Issues Paper 2026/061)

International Monetary Fund

IMF Selected Issues Paper applying the exposure-complementarity framework to Israel: 23.1% of workers hold high-exposure, low-complementarity jobs (displacement risk), 32.8% are in low-exposure occupations, and most workers stand to benefit. Highest risk: business/administration professionals, sales workers, ICT specialists. Policy conclusion: comprehensive lifelong learning strategy.

Aggregate Gains from AI and Their Distribution: Global Evidence from Usage Data (WP 2026/147)

IMF

First multi-wave cross-country estimate of aggregate AI gains from usage data: $2.7T/year labor cost equivalent (3.4% of 86-country GDP), with an AI concentration index showing gains tilt to high-wage occupations everywhere — near-total concentration in low-income countries.

Generative AI and labour markets in ASEAN: Significant exposure, limited disruption, uneven preparedness

ILO

ILO brief (8 July 2026) estimating that 22.9% of ASEAN employment (~80 million workers) is in occupations with more than minimal generative AI exposure, while only 3.3% (11.7m) sit in the highest exposure band. Employment in exposed occupations grew from 66m (2017) to 80m (2025), showing no widespread disruption. Women 4.8% vs men 2.3% in the top band. Adoption is concentrated in large firms and technology-intensive occupations, with Singapore MOM data showing 71.5% of firms had not begun AI adoption. Concludes outcomes depend on preparedness and policy choices.

인공지능(AI) 시대 청년 일자리, 직업훈련 확대로 돌파구 찾는다

Ministry of Employment and Labor

Press release from Korea's Ministry of Employment and Labor on a July 8, 2026 roundtable in Jongno, Seoul, where Minister Kim Young-hoon met trainees and graduates of an AI vocational training program run by KG ICT under K-Digital Training (KDT). The ministry describes newly created KDT "AI Campus" specialized courses and the launch of "AI Worker" training as its response to youth employment pressure in the AI era. KG ICT's courses include a physical-AI manufacturing AX campus track that began in June 2026 and an AI battery campus track starting in October 2026; the release cites an application ratio of 3 to 1.

International comparisons show AI effect on productivity

Federal Reserve Bank of Dallas

US productivity grew 2.4% annualized since 2024Q1 (vs 1.6% pre-pandemic); the three most AI-exposed sectors (information, finance/insurance, professional & technical services) grew 3.7% vs 1.7% for the rest, and account for 40% of US productivity gains with only 16% of hours worked. In the EU, where the same exposure index applies, there is no correlation between sector AI exposure and productivity growth. The Anthropic AI Usage Index (US 3.69, EU-27 1.85, Estonia 3.05, Poland 1.41) explains the gap: the exposure-productivity slope strengthens with national AI usage. Correlation, not causation.

Skills England Annual Skills Report 2026

Skills England (GOV.UK)

Skills England's first Annual Skills Report (6 July 2026) estimates 70% of UK workers are in AI-exposed occupations, with highest exposure among professional, analytical and higher-paid cognitive roles. Graduate online job adverts fell 45% in 2025 and entry-level adverts 25%; AI-exposed firms cut junior employment by 5.8%. AI-skilled workers earn a 56% wage premium. Priority occupations (22.9% of the workforce, 7.6M) are projected to add 1.8M jobs by 2035, led by care workers, software developers and IT analysts. The core constraint is skills: business training spend fell 9% (£49.4bn to £44.8bn, 2019-2024), and the government targets upskilling 10M workers by 2030 via TechFirst (£187M) and AI Skills Boost.

Integrating AI in TVET: A practical guide for institutions

UNESCO-UNEVOC

UNESCO-UNEVOC guide (Yang Congkun, Wu Wenxi, Hannes Tegelbeckers; with Shenzhen Polytechnic University and Otto von Guericke University Magdeburg), published 2026-07-03, launched by webinar 2026-07-06, English only. Builds on Beijing Consensus 2019, Ethics of AI Recommendation 2021, GenAI Guidance 2023, AI Competency Frameworks 2024. Five principles (human agency, equity, pedagogical purposefulness, transparency, accountability/data protection) with a procurement rule: if a tool fails any one, do not proceed. Four ecosystem dimensions and a four-phase integration pathway (diagnosis, early adoption incl. automated marking, pedagogical deepening, ecosystem integration). Seven institutional domains; assessment tasks classified AI-excluded/permitted/integrated/enabled; AI-resilient methods; safety-critical occupations keep hands-on competency and human oversight. Launch deck opening figures: 62% of youth used AI daily in 2025 vs 30% formally taught; <10% of institutions follow formal guidance. Full PDF (unesdoc) not readable this week; post based on landing/launch pages, agenda and 33-slide launch deck.

June 2026 Layoff Report: Cuts Cool to 45,849, Down 53% from May, AI Leads Reasons for Fourth Consecutive Month

Challenger, Gray & Christmas

U.S. employers announced 45,849 job cuts in June 2026, down 53% from May. AI was cited in 14,029 cuts (31%), the fourth consecutive month AI led all reasons. YTD 2026 AI-attributed cuts reached 101,743 (~23% of total), versus an estimated 54,836 (5%) for all of 2025. Total YTD cuts were 443,604, down 40% from H1 2025 (744,308). Technology led sectors at 139,156 YTD (up 83% YoY), roughly a third of all cuts. Hiring plans YTD were 91,405, up 10% YoY.

Getting to all-of-the-above: A framework of solutions for AI's coming impacts on work and workers

Brookings Metro

Xavier de Souza Briggs argues there is no silver-bullet policy for AI job disruption and sorts solutions into four families — brakes (human-in-the-loop rules, automation impact assessments), steers (incentives for pro-worker AI design), buffers (modernized UI, wage insurance, right-to-retraining), and shifts (shorter workweeks, wealth-sharing, tax rebalancing). Key finding: "the great mismatch" — AI exposure is inversely correlated with union membership across 1,000+ occupations. Cites private-sector unionization just under 6% (BLS), 55% of business leaders and 68% of investors expecting less entry-level hiring, Meta's 700 + 8,000 AI-pivot layoffs, and state action in Illinois, California, New Jersey and New York. Warns that "training or retraining alone does not produce good jobs."

Annual Economic Report 2026, Chapter I: Progress and peril

Bank for International Settlements

BIS AER 2026 Ch I finds task-level AI time savings of 20-50%, aggregate productivity lift under 1%, US high-exposure sectors showing productivity gains at the expense of lower employment growth, rising earnings-call signals of labour-input reduction, and warns AI investment (>$1tn hyperscaler capex 2025-26) resembles historic booms with sustainability risk.

Anthropic Economic Index report: Cadences

Anthropic

Anthropic Economic Index June 2026 report "Cadences" analyzing how AI usage mirrors human rhythms — workweeks, daily schedules, key dates. Based on privacy-preserving telemetry (Apr 10–Jun 10, 2026) plus a linked survey of ~9,700 respondents. Key findings: personal use rises from ~35% weekday to ~50% weekend; tax queries spiked 8x on April 14; 93% of conversations produce artifacts; higher-wage occupations (marketing managers, programmers) use 2.5x more tokens and 1.34x output per turn; over 35% expect AI to handle most tasks within 12 months yet workers delegating most are the most optimistic (86% productivity gains, only 10% fear job loss); early-career workers carry the most concern (1/3 see junior job-loss probability >60%); computer/math roles 30% of respondents vs 4% of US employment; physical occupations underrepresented. Underlying data: BLS OEWS (May 2025), World Bank WDI, UN WPP.

What separates firms that use AI intensively from firms that don't?

European Central Bank

ECB SAFE survey round 37 (Oct-Dec 2025, 5,000+ euro area firms): 70%+ of firms use AI but only 7% intensively. Intensive users are smaller, younger, ICT/professional-services firms driven by growth not cost-cutting; 84% have completed AI investments and allocate ~20% of total investment to AI. Top barriers: skills shortage (40%) and finance.

Human Capital, AI, and Labor Commoditization

arXiv

Study of 49,610 Upwork workers and 2.26M contracts (2021Q1-2026Q1) finds generative AI reduced the importance of human capital signals (credentials, experience, reputation) by 7.8% in AI-exposed categories. Contract volume fell ~7%, demand shifted to lower-cost workers (3.2%->7.9% share), price weight rose (1.1%->1.8%), and the high vs low human-capital demand gap narrowed (10.3%->6.2%). Average AI exposure across 102 subcategories was 0.252; highest Legal Translation (0.80), lowest Photography (0.02).

Workplace artificial intelligence use: A profile of sociodemographic and job characteristics

Statistics Canada

StatCan survey: generative AI use among Canadian workers nearly doubled from 17% to 30% in under a year (Sep 2024–Jul 2025), concentrated in degree-holding analytical and knowledge-sector roles.

Agentic coding and persistent returns to expertise

Anthropic

Analysis of ~400,000 Claude Code sessions (Oct 2025-Apr 2026, ~235,000 users): users make ~70% of planning but only ~20% of execution decisions. Verified success 15% (novice) vs 28-33% (intermediate/expert); troubled-session recovery 4% vs 15%. Software occupations 34% vs other occupations 29%. Debugging share fell 33%->19%; average task value +27%.

PwC 2026 Global AI Jobs Barometer: Two futures for jobs in an AI era

PwC

PwC analyzed 1B+ job ads across 27 countries. AI-exposed firms grew headcount 52% vs 36% and wages 24% vs 17% since 2018. AI skills wage premium hit 62% (up from 57%). Two-track market: professionalised jobs grow 2x faster with 42% faster wage growth than democratised jobs. AI-exposed entry-level roles up 35% since 2019 while others fell 10% ("seniorization"); junior AI roles 7x more likely to require senior skills. AI-skill jobs grow 8x faster (69% vs 9%). Tech/media/telecom led at 11%.

Entry-level jobs support: AI bootcamps and tech training as government supports young people into the jobs of the future

UK Government (DSIT & DWP)

The UK launched a £20m Early Careers Jobs Alliance (co-chaired by the Prospect union, with partners including Microsoft, Accenture, BAE Systems and JD Sports) alongside the TechFirst programme reaching 400,000 students, an £820m Youth Guarantee supporting almost 1 million young people via 350,000 placements and 360+ youth hubs, and a phased AI bootcamps rollout (North West summer 2026, nationwide 2027-2028). National target: 10 million UK workers AI-upskilled by 2030 (1.7m courses completed).

Climbing the employment ladder tough when bottom rung is broken

Federal Reserve Bank of Dallas

Dallas Fed economists find youth (under-25) employment in highly AI-exposed occupations fell ~13% since 2022. The decline comes not from layoffs but from a falling job-finding rate — young workers cannot get hired in the first place. Firms pull back on entry-level recruitment while competing aggressively for proven, experienced talent, because AI now handles much codified (textbook) knowledge while tacit (experience-based) knowledge stays scarce. Computer systems design wages rose 16.7% since fall 2022 vs 7.5% nationally, evidence of a widening experience premium.

Texas Business Outlook Surveys: Special Questions (AI), May 2026

Federal Reserve Bank of Dallas

Survey of 313 Texas executives (May 12-20, 2026). 66.2% of firms use AI (flat vs Dec 2025). Among 203 AI-using firms, 76.4% report no impact on their need for workers; only 10.4% report any decrease. But 25.7% expect a decrease over the next few years (29.3% among current users) - a ~2.5x gap between realized and expected cuts. 71.4% report higher productivity for AI-using employees, yet only 7.4% report higher wage growth for them.

Generative AI and the Reorganization of Labor Demand

arXiv

Wang, Wei & Wang (2026) measure AI exposure dynamically at the job-posting level across the U.S. economy using a two-stage LLM pipeline. Decomposing the decline in AI exposure: 52% comes from hiring reallocation, 39.5% from within-job task redesign; observable job characteristics explain ~90% of exposure changes (Oaxaca-Blinder). Senior roles adjust earlier via reallocation; junior roles via a broader mix of reallocation, redesign, and their interaction.

The adoption of AI by industrial sectors

PIIE

PIIE analysis by Hufbauer & Zhang finds AI adoption is skill-biased: high-wage service sectors (information, finance, professional services) lead with 30%+ firm adoption while retail/hospitality/food services lag. Large firms (250+) tripled adoption from ~4% to ~12% (2023-2025); large-firm employment share rose 37.6% (2000) to 42.3% (2025). Modest positive correlation between compensation and AI adoption. Labor displacement effects deemed premature/inconclusive.

How ChatGPT adoption broadened in early 2026

OpenAI Signals Research

OpenAI Signals Q1 2026 update documents ChatGPT consumer adoption broadening across age (35+ gaining share), gender (feminine-name users now majority), and geography (Dominican Republic +9, Haiti +9, Japan +8, Mexico +6, Tanzania +6, Brazil/Costa Rica/Myanmar/Papua New Guinea +5 in per-capita messaging rank between Q4 2025 and Q1 2026).

April 2026 Job Cuts Report: Cuts Rise 38% From March, YTD Cuts Down 50%

Challenger, Gray & Christmas

April 2026: 83,387 US job cuts (+38% MoM, -21% YoY). AI #1 reason at 21,490 cuts (26%), second straight month. YTD AI cuts 49,135 (16% of all 2026 cuts, up from 13%). YTD total 300,749 (-50% YoY). Technology led industries with 33,361 cuts in April (85,411 YTD, +33%). Hiring announcements down 69% MoM to 10,049.

AI growth acceleration versus distributional fairness

Brookings Institution

Brookings synthesis arguing AI-driven growth acceleration is plausible but not guaranteed. Diffusion is highly uneven across sectors, firms, countries, demographic groups. Relative employment declines for early-career workers in AI-exposed occupations identified using high-frequency payroll data. Five-pillar policy recommendation including complementary capabilities investment, strategic procurement, transition tools, measurement standards, broad-based asset ownership.

Artificial Intelligence in German establishments: One in four has adopted generative AI (IAB-Kurzbericht 08/2026)

IAB (Institute for Employment Research)

IAB-Kurzbericht 08/2026 (Friedrich & Kagerl) reports that generative AI adoption among German establishments rose nearly fivefold from 5% (2023) to 24% (2025), based on the ~15,000-establishment IAB-Betriebspanel. Adoption is 48% at firms with 200+ employees vs 21% at firms under 10; leading sectors are information & communication (59%) and finance & insurance (50%), while construction, hospitality, transport and raw materials stay below 15%. 90% of adopters use off-the-shelf tools; only 16% train purchased models and 6% build their own. About half invested financially, 25%+ offer AI training, and 20% have workplace AI rules.

What Jobs Can AI Learn? Measuring Exposure by Reinforcement Learning

arXiv

Tomei & Klein Teeselink introduce the RL Feasibility Index (RLFI), scoring all 17,951 O*NET tasks across 894 occupations on whether reinforcement learning could learn them. A physical-feasibility gate zeroes 40.7% of tasks. RLFI correlates 0.88 overall with Eloundou LLM exposure but only 0.15 among digitally feasible tasks — diverging on which reachable jobs AI can learn. High RLFI/low LLM-exposure: power plant operators, railroad conductors, aircraft cargo handling supervisors. Low RLFI/high LLM-exposure: musicians, physicians, natural sciences managers. RL exposure is hump-shaped by wage (peaks upper-middle, +12.2pt per log-salary point) and inverted-U by seniority (mid-career peak). Post-ChatGPT DiD: +1SD RL exposure ~ -2.9% job openings (p=0.085).

Evaluating the Impact of AI on the Labor Market: March 2026 CPS Update

Yale Budget Lab

March 2026 CPS update finds no meaningful shift from December across most categories. Occupational dissimilarity, industry dissimilarity, and exposure/usage metrics all remain flat or continue along existing trends. Notable uptick in occupational mix dissimilarity between older and younger college graduates (high end of historical range). Anthropic February 2026 usage data shows observed usage more associated with automation than augmentation, consistent with prior November 2025 release. Conclusion: anxiety widespread but data suggests AI labor market impact remains largely speculative — picture reflects stability, not major disruption at economy-wide level.

Today at Work 2026, Issue 1: AI Powers Into the Workplace

ADP Research

ADP 2025 Global Workforce Survey (payroll data for 25M+ US workers + 600,000+ workers across 34 countries) finds 50% of workers use AI at least a few times weekly and 20% almost daily. Daily AI users report higher full engagement (30% vs 14% non-users; 19% baseline) and lower overload (11% vs 23%), yet are 4x more likely to feel less productive than capable — a productivity paradox the authors attribute to workers perceiving they personally did less. Young workers and men lead adoption; early-career workers aged 22-25 in AI-exposed fields (software development, customer support) face employment-growth concerns and are less optimistic about job security.

From Clerks to Agentic AI: How Will Technology Transform the Labor Market in Finance?

arXiv

Yu & Li (April 2026) track 40 years of financial labor productivity across three regimes: computerization (1.36 AUM/employee coefficient), indexing (2.42), and agentic AI era (3.39 — 149% higher than baseline). AI exposure correlates +0.5843 with AUM/employee but -0.0535 with revenue/employee, indicating volume-over-yield strategy. Mid-layer coordination roles face greatest pressure.

Workers exposure to AI: What indicators tell us and what they don't

ILO

ILO Research Brief warning that AI exposure indicators (3.3% global highest exposure, 34% HIC vs 11% LIC, female 4.7% vs male 2.4%) measure technical task overlap with current AI capabilities — not predictions of job displacement.

WorkRB: A Community-Driven Evaluation Framework for AI in the Work Domain

arXiv

Evaluation framework for AI work-domain capabilities. Low blog value (framework paper, not labor market impact evidence).

The AI Jobs Transition Framework: Mapping AI's Near-Term Impact on Jobs

OpenAI

OpenAI economist Alex Martin Richmond maps 921 US occupations covering 99.7% of US employment using a 4-dimension framework: technical capability, human necessity (regulatory/relational/physical), demand elasticity, and ChatGPT consumer usage data (H2 2025). Findings: 18% face higher short-term automation risk (legal, education, office/administrative); 46% likely see less change; 24% may see employment decline as task composition shifts; 12% could grow because of AI. Insulated roles include teachers, nurses, lawyers due to regulatory, relational and physical necessity. ChatGPT usage 3x higher in high-risk jobs.

ChatGPT Usage and Adoption Patterns at Work

OpenAI

OpenAI April 2026 workplace report: 28% of US workers use ChatGPT (up from 8% in 2024), Fortune 500 adoption 93%, 7M+ Enterprise/Team seats. Four core tasks: writing, research, programming, analysis. Adoption split sharply between IT/finance (deep) and healthcare/construction/transportation (lagging).

AI Is Probably Not (Yet) the Reason for Labor Market Weakening

The Budget Lab at Yale

Yale Budget Lab March 2026 CPS analysis: US unemployment rose to 4.3% but AI exposure/automation/augmentation metrics show no relationship to employment changes. Immigration slowdown, not AI, is the primary driver of labor market weakening. Dissimilarity index flat since ChatGPT 2022; recent graduates (20-24) show no growing divergence from older workers (25-34).

Inside the AI Index: 12 Takeaways from the 2026 Report

Stanford HAI

Annual AI Index 2026 report highlights: software developer employment (22-25) down ~20% since 2024, 14-26% productivity gains in customer support/SW dev, 73% expert vs 23% public optimism gap, US ranks 24th in AI adoption (28.3%), AI researcher US inflow down 89% since 2017, GenAI consumer surplus $172B, 53% global usage, physician note-writing time reduced 83%.

Is your job safe?

ADP Research

ADP Research Global Workforce Survey of 39,000+ workers found only 25% globally and 28% in the U.S. feel their jobs are safe. Age breakdown: 18-26: 26%, 27-39 & 40-54: 30%, 55-64: 23%. Three contributing factors: geopolitical concerns, AI impact, ongoing inflation. Workers feeling secure are 6× more likely to be fully engaged and 3.3× more likely to report high productivity. Authored by Nela Richardson Ph.D. April 7 2026.

AI Impact on Labor: Double-Edged Sword

Morgan Stanley

Morgan Stanley analysis: AI added at most 0.1 percentage point to overall US unemployment rate. Gen AI impact is double-edged — same technology that automates tasks also augments labor and raises productivity.

AI Eliminating 16,000 U.S. Jobs Per Month

Goldman Sachs

Goldman Sachs economist Elsie Peng finds AI substituting 25K jobs/month while augmenting 9K, net loss 16K. Gen Z hardest hit with 3.3pp wage gap widening. Insurance clerks, bill collectors at highest substitution risk.

Challenger Report: March Cuts Rise 25% From February, AI Leads Reasons

Challenger, Gray & Christmas

March 2026: 60,620 total job cuts, AI led all reasons with 15,341 cuts (25%). Q1 YTD: 217,362 total, 27,645 AI (13%). Cumulative AI cuts since 2023: 107,094. Tech sector Q1 up 40%, Healthcare record Q1 high.

How AI may reshape career pathways to better jobs

Brookings Institution

Analysis of 70M STARs (non-degree workers): 15.6M in high AI-exposed roles, 3.5M at dual risk (high exposure + low adaptive capacity). Nearly half of career advancement pathways from Gateway to Destination occupations are highly AI-exposed. Florida metros most affected. Uses Anthropic observed exposure measure.

The 2026 AI Index Report

Stanford HAI

Stanford HAI's 2026 AI Index documents accelerating AI capability, investment and adoption, and highlights early labor-market signals including employment declines among younger workers in AI-exposed entry-level occupations.

Crashing Waves vs. Rising Tides: Preliminary Findings on AI Automation from Thousands of Worker Evaluations of Labor Market Tasks

MIT / arXiv

MIT researchers evaluated 3,000+ text-based O*NET tasks through 17,000+ worker assessments. Found "rising tides" pattern: AI task success climbing from ~50% (2024-Q2) to ~65% (2025-Q3), projected 80-95% by 2029. Little evidence of sudden "crashing waves" displacement. Adoption timelines expected to substantially lag capability gains.

AI and the US Labour Market: Effects on Employment Growth

European Central Bank

ECB Economic Bulletin (Issue 4/2026) finds US occupations at high AI-substitution risk declined more than 4% from 2019-2025 while low-risk occupations grew 13%, a ~15pp employment wedge that widened fastest after ChatGPT's late-2022 launch. Low-risk share rose 23%->25% of employment, high-risk fell 35%->33%. No significant wage-growth effect detected. Difference-in-differences with sector fixed effects.

2026 AI Index Report: Economy Chapter

Stanford HAI

Economy chapter of 2026 AI Index. Key findings: software developers aged 22-25 employment fell nearly 20% from 2024; ~1/3 of surveyed organizations anticipate workforce reductions over coming year; expected decreases concentrated in service operations, supply chain, and software engineering; productivity gains measured at 14-15% (customer support) and 26% (software development). AI labor effects concentrated in hiring pipelines and youngest workers in exposed occupations.

Forecasting the Economic Effects of AI

National Bureau of Economic Research

Karger, Kuusela, Tetlock et al. (15 co-authors) ran a structured forecasting exercise across 5 groups (academic economists, AI company employees, policy researchers, superforecasters, general public). Median GDP growth forecast through 2050: 2.5% (vs CBO baseline 2.0%/1.7%). Under rapid AI scenario: GDP ~4%/year, labor force participation drops 62.7% to 55%, ~10M jobs displaced by AI specifically. Main driver of disagreement: not AI capability timelines, but what high-capability AI does to the economy. Policy preferences: experts favor targeted retraining + portable benefits; general public prefers UBI + federal job guarantees.

Economics of Human and AI Collaboration: When is Partial Automation More Attractive Than Full Automation?

arXiv

MIT/IBM working paper develops a unified framework for optimal task automation. Convex AI accuracy cost curve makes partial automation the cost-minimizing equilibrium. 11% of computer-vision-exposed labor compensation captured at firm level; share rises sharply under economy-wide deployment. Calibrated with O*NET task data, 3,778 domain expert surveys, GPT-4o decompositions.

Agentic AI and Occupational Displacement: A Multi-Regional Task Exposure Analysis of Emerging Labor Market Disruption

arXiv

Introduces Agentic Task Exposure (ATE) score framework. Analyzes 236 info-intensive occupations across 5 US tech hubs. 93.2% cross moderate-risk threshold by 2030 in Tier 1 regions. High-risk: credit analysts, judges, sustainability specialists (ATE 0.43-0.47). 17 emerging occupational categories in AI governance. Authors: Gupta & Kumar.

New ILO-World Bank paper highlights uneven global impact of generative AI on jobs

ILO / World Bank

Joint ILO-World Bank background study for WDR 2026 analyzing GenAI labor market exposure across 135 countries. Finds uneven impact: advanced economies face higher exposure in clerical/professional roles; developing economies risk faster disruption than productivity gains due to digital gaps. Women and youth disproportionately vulnerable.

The AI Layoff Trap: Automation Arms Race

arXiv (Wharton)

ورقة وارتون على arXiv: فخ تسريحات الذكاء الاصطناعي — الشركات المتسابقة نحو الأتمتة تخلق مشكلة عمل جماعي قد تأتي بنتائج عكسية.

A people-first vision for the future of work in the age of AI

Brookings Institution

Policy vision by Friedler, Booth, Schrank, Helper proposing worker-centered AI policies: minimum staffing laws in care sectors, tripartite institutions, participatory AI design. Cites 50%+ Americans fear AI displacement (Reuters/Ipsos). Occupations mentioned: teachers, nurses, social workers, software engineers, manufacturing, utility, guest room attendants.

Economic Index March 2026: Learning Curves

Anthropic

مؤشر أنثروبيك الاقتصادي مارس 2026: 49% من الوظائف تستخدم الذكاء الاصطناعي في أكثر من 25% من المهام. منحنيات التعلم تظهر تقدم المتبنين الأوائل.

How AI Is Transforming Work at Anthropic (Internal Data)

Anthropic

بيانات أنثروبيك الداخلية: المهندسون يستخدمون الذكاء الاصطناعي في 59% من عملهم. تكشف عن أنماط الاستخدام الفعلية مقارنة بالتقديرات الخارجية.

AI Hiring Is Booming While Everything Else Stalls

Stanford HAI + Indeed

بيانات ستانفورد HAI + Indeed: التوظيف في مجال الذكاء الاصطناعي يرتفع بينما يتوقف التوظيف في غيره، مما يخلق سوق عمل بمسارين.

Building Pro-Worker AI

Brookings (Acemoglu, Autor, Johnson)

أسيموغلو وأوتور وجونسون (بروكينغز): ثلاثة اقتصاديين بارزين يرون أن المسار الحالي لتطوير الذكاء الاصطناعي يضر بالعمال.

AI and the Entry-Level Job

Harvard Business Review + IBM

IBM تضاعف توظيف المبتدئين ثلاث مرات بينما يخفض المنافسون. دراسة حالة HBR عن استراتيجية قوى عاملة مخالفة للتيار في عصر الذكاء الاصطناعي.

Gen AI Wont Make Your Employees Experts

Stanford + Harvard

دراسة ستانفورد-هارفارد: الذكاء الاصطناعي لا يستطيع سد فجوات الخبرة. دراسة ميدانية لمدة 8 أشهر تظهر أن الذكاء الاصطناعي يعزز الفوارق القائمة في المهارات.

How AI Is Changing the Labor Market

Dallas Federal Reserve + HBR

تحليل HBR + الاحتياطي الفيدرالي في دالاس: الذكاء الاصطناعي يساعد ويضر العمال في آن واحد عبر قنوات مختلفة — الإنتاجية مقابل الإزاحة.

Equipping workers with insights about compensation

OpenAI + University of Michigan

ChatGPT wage usage study: In Jan-Feb 2026, US consumer ChatGPT users sent ~3M messages/day about wages, compensation, or earnings. Workers use ChatGPT to navigate wage info gaps, especially in labor markets with high uncertainty. Joint study with University of Michigan on how Americans engage with AI for wage information. Released 2026-03-17.

AI and Young Workers: Employment Effects

Dallas Federal Reserve

بيانات الاحتياطي الفيدرالي في دالاس تظهر انخفاض توظيف الشباب في قطاع التكنولوجيا منذ أوائل 2022، أي قبل أشهر من إطلاق ChatGPT.

Disruption without dividend? How the digital divide and task differences split GenAI's global impact

International Labour Organization

ILO Working Paper 166 examines generative AI's uneven labor-market effects across 135 countries. Advanced economies face higher automation exposure (30-32% of employment), but developing nations face a paradox: workers vulnerable to displacement already have internet connectivity, while those positioned to gain from AI-driven productivity often lack reliable digital infrastructure. The paper warns that "disruptive effects may materialize before benefits" in some developing economies, underscoring gaps between AI's technological potential and actual implementation capacity.

New Data Show No AI Jobs Apocalypse — For Now

Brookings Institution

تحليل بروكينغز على مدى 33 شهراً: لا بطالة جماعية بسبب الذكاء الاصطناعي حتى الآن، لكن هناك علامات تحذيرية ناشئة في قطاعات محددة.

Evaluating the Impact of AI on the Labor Market: January/February CPS Update

Yale Budget Lab

33 months after ChatGPT release, AI exposure metrics, occupational dissimilarity, and industry dissimilarity all flat or within historical ranges. Recent vs older college graduate dissimilarity in 30-33% band since Jan 2021. Anthropic Feb 2026 usage data tilts toward automation rather than augmentation. Direct quote: exposure/automation/augmentation measures show no sign of being related to changes in employment or unemployment.

Generative AI and Jobs: Refined Global Index of Occupational Exposure

ILO

مؤشر منظمة العمل الدولية المحسّن عبر 138 دولة: النساء يواجهن ضعف خطر الأتمتة بسبب أنماط الفصل المهني.

AI Labor Displacement and the Limits of Worker Retraining

Brookings Institution

يدرس بروكينغز برامج إعادة التدريب التاريخية ويجد معدلات نجاح محدودة للعمال المُزاحين من وظائفهم.

World Employment and Social Outlook: Trends 2026

ILO

تقرير رئيسي لمنظمة العمل الدولية: البطالة العالمية تبلغ 186 مليون شخص، ومفارقة الذكاء الاصطناعي تتجلى في مكاسب إنتاجية دون تعافٍ وظيفي. 408 ملايين يفتقرون إلى عمل لائق.

Research on AI and the Labor Market Is Still in the First Inning

Brookings Institution

تحليل بروكينغز الشامل يرى أن أبحاث تأثير الذكاء الاصطناعي على سوق العمل لا تزال في مراحلها الأولى — معظم الادعاءات تسبق الأدلة المتاحة.

Karpathy AI Job Exposure Score for Every US Job

Andrej Karpathy / Fortune

كارباثي (من OpenAI سابقاً) يقيّم كل وظيفة أمريكية حسب التعرض للذكاء الاصطناعي. المهنيون ذوو الياقات البيضاء يظهرون أعلى درجات الهشاشة.

Measuring US Workers Capacity to Adapt to AI-Driven Job Displacement

Brookings Institution

بروكينغز يحدد 6.1 مليون عامل أمريكي معرضين بشكل كبير للذكاء الاصطناعي مع قدرة منخفضة على التكيف — خريطة الهشاشة الهيكلية.

Labor market impacts of AI: A new measure and early evidence

Anthropic Economic Research

Anthropic introduces "observed exposure" — a new metric measuring AI's actual deployment in real workflows. Computer Programmers top at 75% observed exposure; full Computer & Math category sits at 33% (vs 90% theoretical). 30% of workers have zero observed exposure. New hires aged 22-25 in high-exposure fields drop ~0.5pp.

AI and Young-Adult Jobs: The Real Mystery

Economic Innovation Group

دراسة EIG تتحدى رواية تأثير الذكاء الاصطناعي على وظائف الشباب: انخفاض التوظيف له أسباب غير مرتبطة بالذكاء الاصطناعي تشمل تصحيحات ما بعد الجائحة.

Reverse gear: how AI is bringing vocational occupations back

Cedefop

Cedefop analysis of EU online job advertisements shows VET (vocational education and training) occupations' share fell from 36% (2019-mid 2022) to 33% (mid 2022 trough), then recovered above 36% by early 2025 — coinciding with the emergence of generative AI in late 2022. Declining categories: software developers, sales & marketing, customer info workers, database specialists. Growing categories: engineering technicians, machinery mechanics, construction trades, transport workers.

Tracking the Impact of AI on the Labor Market

Yale Budget Lab

The Yale Budget Lab tracks AI's labor-market impact using current population data and finds that occupational exposure has stayed broadly stable since the launch of ChatGPT, with no clear sign of large aggregate disruption to employment yet.

AI, Productivity, and the Workforce: Evidence from Corporate Executives

NBER

Survey of 750 corporate executives. Over half invested in AI. Little evidence of near-term aggregate employment declines. Larger companies anticipate reductions, smaller firms expect gains. Productivity paradox: executives perceive larger gains than data shows.

Generative AI, Occupational Segregation and Gender Equality in the World of Work

ILO

This ILO study examines how generative AI intersects with occupational gender segregation, concluding that women are roughly twice as likely as men to be in highly exposed clerical roles, which could widen labor-market inequalities without targeted policy.

Generative AI and Jobs: A Refined Global Index of Occupational Exposure (Research Brief)

ILO

This ILO research brief presents a refined global index of occupational exposure to generative AI, finding that clerical and administrative jobs are most exposed and that women face higher exposure than men because of their concentration in such roles.

India Brief: Economic Index

Anthropic

موجز أنثروبيك عن الهند: ثاني أكبر قاعدة مستخدمين للذكاء الاصطناعي عالمياً لكنها في المرتبة 101 من حيث نصيب الفرد — فجوة الحجم مقابل العمق.

Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI

NBER

Danish admin data + adoption surveys show rapid AI chatbot adoption and reported productivity gains, but precise null effects on earnings and recorded hours at worker/workplace level (within 2%) two years after ChatGPT launch. Employers reorganized tasks rather than reducing workforce.

February 2026 Job Cuts Report

Challenger, Gray & Christmas

تقرير تشالنجر: 12,304 تسريحات مرتبطة بالذكاء الاصطناعي في يناير-فبراير 2026، بينما انخفضت خطط التوظيف بنسبة 56% مقارنة بالعام السابق.

Fed's Barr Names 3 AI Scenarios — Including 'Essentially Unemployable'

Bank for International Settlements (BIS Review)

AI Doesnt Reduce Work — It Intensifies It

Harvard Business Review

دراسة HBR الميدانية لمدة 8 أشهر تكشف أن الذكاء الاصطناعي يزيد كثافة العمل بدلاً من تقليل عبئه — مفارقة الإنتاجية.

An AI Productivity Boom? Don't Count Your (Productivity Data) Chickens

Yale Budget Lab

Skeptical analysis of AI productivity boom claims. Productivity data is too noisy and lagged to confirm GenAI productivity gains.

Is Generative AI a Job Killer? Evidence From the Freelance Market

Brookings Institution

دراسة بروكينغز لسوق العمل الحر: المستقلون ذوو المهارات العالية هم الأكثر تضرراً من الذكاء الاصطناعي، مما يناقض فرضية إزاحة العمالة منخفضة المهارة.

Yale Budget Lab: AI Exposure Hasn't Moved Workforce Distribution Since ChatGPT

The Budget Lab at Yale

Firm Data on AI

NBER

Survey of 6,000 executives across US/UK/Germany/Australia finds 69% use AI but 90% report zero employment/productivity impact. Executives predict -0.7% employment next 3 years; employees expect +0.5% growth.

Building Pro-Worker AI

The Hamilton Project (Brookings)

Acemoglu, Autor and Johnson argue that AI should be steered toward complementing workers rather than replacing them, identifying barriers such as misaligned incentives and a pro-automation bias, and proposing nine policy measures to make AI more pro-worker.

Payrolls to Prompts: Firms Replacing Labor With AI

arXiv (Payrolls to Prompts)

دراسة arXiv: الشركات تنفق 3 سنتات فقط على الذكاء الاصطناعي مقابل كل دولار تخفضه من ميزانيات المستقلين — الاستبدال غير مكتمل.

Graph-Based Analysis of AI-Driven Labor Market Transitions

arXiv

Analyzes 9,978 Egyptian job postings using knowledge graphs. Finds 20.9% of jobs face high automation risk, but only 24.4% of at-risk workers have viable transition pathways. 75.6% face structural barriers requiring comprehensive reskilling, not incremental upskilling. Process-oriented skills appear in 15.6% of feasible transitions.

[제2026-3호] '쉬었음' 청년층의 특징 및 평가: 미취업 유형별 비교 분석 (Characteristics and Assessment of 'Resting' Youth)

Bank of Korea

Analysis of Korean youth in "resting" status: 6.3%p higher probability for non-college youth, reservation wage 31M KRW (not high), AI cited as labor market deterioration factor, debunks "high standards" narrative

Companies Are Laying Off Workers Because of AIs Potential, Not Performance

Harvard Business Review

تحليل HBR: المدراء التنفيذيون يسرّحون العمال بناءً على إمكانات الذكاء الاصطناعي وليس أدائه الفعلي. استثمار واحد فقط من كل 50 في الذكاء الاصطناعي يحقق تحولاً حقيقياً.

How AI is changing early careers: A view from entry-level workers

PwC + World Economic Forum

PwC+WEF joint survey of 9,394 entry-level workers across 48 economies. 47% curious about AI, 38% excited, 29% worried. 76% say job security is most important factor, but only 53% feel confident. Presented at Davos 2026.

Payrolls to Prompts: Firm-Level Evidence on the Substitution of Labor for AI

arXiv

Ryan Stevens (Ramp applied science) analyzes firm-level spending on Ramp expense management platform (2021 Q3 - 2025 Q3). Top-exposed firms substitute $1 in labor for only $0.03 in AI spend (33x cost savings); mid-exposed firms at $0.30. Online labor market spending fell from 0.66% (2021 Q4) to 0.14% (2025 Q3); AI spending rose from 0% to 2.85% over same period. 50%+ of firms reduced online labor spending to 0%. Limitation: covers online freelance markets only, no occupation-level decomposition.

Anthropic Economic Index: January 2026

Anthropic

مؤشر أنثروبيك الاقتصادي: مكاسب الإنتاجية الفعلية من الذكاء الاصطناعي هي 1.0% فقط وليس 1.8% المعلنة — معظم الاستخدام تعزيز وليس أتمتة.

Looking for the Ladder: Is AI Impacting Entry-Level Jobs?

Economic Innovation Group

In this EIG working paper, Zanna Iscenko and Fabien Curto Millet (Google economics team) analyze whether AI is eroding entry-level employment. They find job postings declined more in AI-exposed occupations, but the inflection point began in 2022, months before ChatGPT's public release. The timing aligns better with the macroeconomic shift of rising interest rates than with the launch of large language models, challenging the dominant narrative that AI is the primary driver of weakening entry-level hiring.

AI-Exposed Jobs Deteriorated Before ChatGPT

University of Pittsburgh + RAND

دراسة بيتسبرغ/RAND على 10.5 مليون ملف LinkedIn: الوظائف المعرضة للذكاء الاصطناعي تدهورت قبل 8-10 أشهر من إطلاق ChatGPT.

AI is reshaping entry-level hiring while lifting pay for experienced workers

Federal Reserve Bank of Dallas

Dallas Fed economists examine employment and wage trends across more than 200 occupations since ChatGPT's late-2022 release. They find AI is simultaneously reducing employment in the most exposed industries (computer systems design down roughly 5%) while pushing wages higher for experienced workers in those same fields. Entry-level workers face the steepest challenge because AI can replicate codified textbook knowledge but not the tacit knowledge built through years of experience, so the labor-market impact hinges on whether AI automates or augments tasks.

The end of work feels near. How do people perceive the impact of digital technologies and automation?

IAB / Labour Economics

Survey experiment with 5,147 US/German workers (2019, YouGov). Over 50% expect automation to raise aggregate unemployment; ~90% expect unequal impacts; but under 30% fear for their own job (correlation between the two below 0.1). Randomized provision of scientific information (Graetz & Michaels 2018: robots did not significantly reduce overall employment) cut aggregate-unemployment fear by ~0.15 SD, persisting 4 weeks later (n=2,225 follow-up). Effects heterogeneous by prior beliefs, with opposing shifts in policy demand. Pre-generative-AI data.

Enhancing Worker Productivity Without Automating Tasks: A Different Approach to AI and the Task-Based Model

NBER

Agrawal/McHale/Oettl: AI as productivity tool vs task replacer, wage inequality depends on skill distribution

AI 인재 5.7만 명 시대, 왜 기업은 사람이 없다고 할까?

한국은행

Korea has 57K AI specialists but 6% wage premium (vs US 25%), 16% brain drain, 30% firms lack AI job definitions.

Preparing for 2040: four AI-powered scenarios for the future of continuing skills development

Cedefop

Cedefop foresight report naming AI the most influential yet unpredictable force shaping continuing skills development (CSD) by 2040, presenting four scenarios (A opportunities / B polarisation / C slow adoption / D dystopia) and cross-cutting red flags. Paired with the 2024 European AI skills survey (42% need AI skills vs 15% trained = 27pp gap).

Stanford AI Experts Predict What Will Happen in 2026

Stanford HAI

2026 predictions: emergence of high-frequency AI economic dashboards tracking task/occupation-level productivity, displacement, new roles. Shift from speculation to careful measurement. Round-up prediction article, not primary research.

AI Jobs Barometer 2025

PwC

مقياس PwC يكشف أن الوظائف المعرضة للذكاء الاصطناعي تحقق مكاسب إنتاجية 4 أضعاف وعلاوة أجور 56% مقارنة بالوظائف غير المعرضة.

AI 전문인력 현황과 수급 불균형: 규모, 임금, 이동성 분석

한국은행

BOK Issue Note 2025-36: 57K AI workforce with 58% holding graduate degrees, 69% large firms plan to expand AI hiring.

Beyond Automation: Redesigning Jobs with LLMs to Enhance Productivity

arXiv

UK Civil Service study analyzing 193,497 job vacancies and 1,542,411 tasks for AI exposure. Finds productivity gains, not just displacement. Authors propose redesign framework: automate, optimize, reallocate.

Incorporating AI Impacts in BLS Employment Projections 2024-34

Bureau of Labor Statistics

مكتب إحصاءات العمل الأمريكي يدمج تأثيرات الذكاء الاصطناعي في التوقعات الرسمية للتوظيف 2024-2034 لأول مرة — تحول منهجي.

Agents, Robots, and Us: Skill Partnerships in the Age of AI

McKinsey Global Institute

MGI report: $2.9T US value by 2030, 57% work hours automatable, 40% jobs highly automatable, 70%+ skills transferable

AI 확산 초기, 청년고용은 왜 감소하는가?

한국은행

Youth employment declined 2.11M over 3 years with 98.6% in AI-exposed sectors. Junior workers with codified knowledge most vulnerable.

The State of AI: Global Survey 2025

McKinsey & Company (QuantumBlack)

McKinsey's QuantumBlack surveyed 1,993 participants across 105 nations (June-July 2025) on enterprise AI adoption. Some 88% of respondents report regular AI use in at least one business function, up from 78% a year earlier, though most enterprises remain in experimenting or piloting stages and only about a third have begun scaling. Sixty-two percent are at least experimenting with AI agents and 23% are scaling agentic systems, yet just 39% report enterprise-level EBIT impact. The highest-value performers redesign workflows and pursue growth and innovation, not efficiency alone.

AI 확산과 청년고용 위축: 연공편향(seniority-biased) 기술변화를 중심으로 [BOK 이슈노트 제2025-30호]

한국은행 (Bank of Korea)

BOK Issue Note No. 2025-30. In Korea over the past 3 years, youth (ages 15-29) employment in high-AI-exposure sectors fell by 211,000 jobs, of which 208,000 (98.6%) occurred specifically in AI-high-exposure sectors. Senior (50+) employment grew by 209,000 in same period, with 146,000 in AI-exposed sectors. AI diffusion exhibits 'seniority-biased technological change' — entry-level (codifiable, routinized) jobs replaced by AI, while jobs requiring tacit knowledge and social skills show complementarity. Korea GenAI adoption rate 63.5% (work-use 51.8%) — roughly 2x US, 8x faster diffusion than internet adoption.

AI and Demographic Changes

Yale Budget Lab

Martha Gimbel analysis: AI exposure intersects with workforce aging and immigration. Some aging-heavy occupations (legal secretaries, admin assistants) show higher AI exposure; passenger attendants show minimal applicability. AI exposure concentrates in occupations with lower shares of foreign-born workers. Commentary piece, no numerical breakdowns in article body.

GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks

OpenAI

GDPval benchmark: 1,320 tasks across 44 knowledge-work occupations in 9 top-GDP sectors (Real estate, Government, Manufacturing, Professional services, Health care, Finance, Retail, Wholesale, Information). Top models (GPT-5, Claude Opus 4.1) rated as good as or better than humans in ~50% of tasks. 100x faster and cheaper than experts. Occupations include: software developers, lawyers, registered nurses, mechanical engineers, accountants, financial analysts, customer service reps, pharmacists, project management specialists, police supervisors, producers/directors, and others. Based on O*NET + BLS May 2024 wage data.

Evaluating the Impact of AI on the Labor Market: Current State of Affairs

The Budget Lab at Yale

This Budget Lab at Yale analysis evaluates whether AI is measurably reshaping the US labor market, using occupational and industry dissimilarity metrics alongside AI-exposure measures. The data show no substantial acceleration in the rate of change in labor-market composition since ChatGPT's introduction; dissimilarity and exposure metrics remain flat or follow prior trends. The authors conclude the picture reflects stability rather than major economy-wide disruption, while cautioning that better data is needed and committing to regular updates of the analysis.

Remote Labor Index: Measuring AI Automation of Remote Work

arXiv

Introduces a Remote Labor Index to measure AI automation capability across remote-work tasks. Key finding: AI capabilities on remote-work benchmarks are rising rapidly but remain below human baseline for complex multi-step tasks. Relevant for occupations dominated by remote knowledge work.

AI and jobs: A review of theory, estimates, and evidence

arXiv

Meta-review of AI exposure and employment literature. Synthesizes theoretical frameworks, empirical estimates, and emerging evidence across 2023-2025. Tests consistency of exposure scores (Felten, Webb, Pizzinelli) with observed hiring patterns.

How People Use ChatGPT (NBER Working Paper 34255)

OpenAI / NBER

Aaron Chatterji, Thomas Cunningham, David J. Deming, Zoe Hitzig, Christopher Ong, Carl Yan Shan, Kevin Wadman. NBER WP 34255 (Sep 2025). ChatGPT adopted by ~10% of world adult population by Jul 2025; gender gap narrowing; higher growth in lower-income countries. Non-work messages grew 53% -> 70%+. 75% of conversations focus on practical guidance, information seeking, and writing. 49% of messages are Asking mode (advisor) showing decision-support value especially in knowledge-intensive jobs. Privacy-preserving automated classifier on de-identified messages. Companion to OpenAI compensation paper (#64) and GDPval (#65).

Yes, AI is affecting employment. Here's the data.

ADP Research

ADP Research examines how generative AI is reshaping employment across worker age groups, drawing on Stanford Digital Economy Lab findings. Employment for workers aged 22-25 fell about 6% between late 2022 and July 2025 in AI-exposed fields like software development and customer support, while older workers in the same roles saw employment gains of 6-13%. The analysis distinguishes AI tools that automate tasks from those that augment human work, concluding that early-career professionals in easily automatable roles face the greatest vulnerability.

[제2025-22호] AI의 빠른 확산과 생산성 효과: 가계조사를 바탕으로 (Rapid Spread of AI and Its Productivity Effects: Evidence from a Household Survey)

Bank of Korea

63.5% of Korean workers use GenAI (51.8% for work), 3.8% work time reduction, 1.0% productivity gain, autonomous robot collaboration 11%->27%, equalizing effect for junior workers

Our Gen AI Transition: Industry Exposures & Adoption (Industry Data)

Jobs and Skills Australia

A companion dataset to Jobs and Skills Australia's Generative AI Capacity Study (14 August 2025), this analysis presents AI exposure and adoption patterns at the industry level across the Australian economy. It complements the occupation-level data by showing which sectors face the greatest AI exposure and where adoption is concentrated, supporting a whole-of-labour-market view of generative AI's potential, its impact to date, and what is needed to support Australia's digital and AI transition.

Our Gen AI Transition: Implications for Work and Skills

Jobs and Skills Australia

JSA Gen AI Capacity Study Overarching Report. First national whole-of-labour-market study by Australian government. 358 occupations analyzed using ANZSCO + two-axis (augmentability + automatability) framework. Key findings: 79% low risk, 21% meaningful exposure (routine clerical concentrated).

Our Gen AI Transition: Exposures, Adaptation, Dynamism (Occupation Data)

Jobs and Skills Australia

Part of Jobs and Skills Australia's Generative AI Capacity Study (released 14 August 2025), this dataset maps AI exposure by occupation using ANZSCO v1.3 unit-level classifications. It measures how jobs are changing in response to AI, including the extent of task adaptation, worker mobility between roles, and overall occupational dynamism. The findings show augmentation generally outweighs automation, with higher automation potential concentrated in routine roles and women-dominated occupations, while entry-level positions show no evidence of widespread displacement in Australia yet.

AI and Jobs: The Final Word (Until the Next One)

Economic Innovation Group

EIG researchers Sarah Eckhardt and Nathan Goldschlag examine whether AI is currently driving meaningful job losses across the US labor market, testing five different AI-exposure measures against unemployment, labor-force participation, occupational switching, and industry employment. They find no substantial employment losses among AI-exposed workers; highly exposed workers actually show lower unemployment than less-exposed peers. The authors conclude AI adoption is accelerating (~9% of businesses) but real economic diffusion remains modest, with most firms reporting neutral or positive employment effects.

Advancing AI Capabilities and Evolving Labor Outcomes

arXiv

Empirical analysis linking AI capability advancement to labor market outcomes. Higher AI exposure associated with reduced employment, higher unemployment rates, and shorter work hours. Evidence consistent with 2025 AI-related hiring slowdowns and layoffs.

OECD Employment Outlook 2025

OECD

The OECD Employment Outlook 2025 reviews labor-market conditions across member countries and examines how AI adoption is affecting jobs, wages and skill demand, finding uneven exposure across occupations and limited net employment effects so far.

AI Index Report 2025

Stanford HAI

مؤشر ستانفورد HAI للذكاء الاصطناعي 2025: تحليل شامل عبر دول متعددة لمعدلات تبني الذكاء الاصطناعي مقابل اتجاهات البطالة.

Augmenting or Automating Labor? The Dual Impact of AI on Jobs and Wages

arXiv

Studies 2015-2022 US data using instrumental variables. Finds automation AI negatively impacts new work, employment, and wages in low-skilled occupations, while augmentation AI fosters new work and raises wages for high-skilled occupations. Concludes AI may accelerate existing wage inequality.

The Future of Jobs Report 2025

World Economic Forum

The WEF Future of Jobs Report 2025 surveys more than 1,000 employers worldwide on how AI and automation will reshape roles by 2030, projecting 92 million jobs displaced alongside 170 million created and a shift in 39% of workers' core skills.

Technological Disruption in the Labor Market

NBER

Deming/Ong/Summers: Historical labor disruption analysis, STEM jobs 50%+ surge since 2010, AI as potential GPT