The Impact of Artificial Intelligence on Jobs

Executive summary

AI has not produced mass unemployment as of October 2026, but young workers in the most exposed jobs are paying a measurable price, and the evidence is still short and mixed. This report uses sources published through July 2026 and was checked on 30 October 2026.

1. Aggregate employment shows little AI effect so far. Yale Budget Lab found no discernible disruption in US employment 33 months after ChatGPT's release (analysis of 1 October 2025). The OECD's June 2026 Economic Outlook found no signs of widespread labour displacement at industry level. S&P Global's firm survey is the exception, with a small net negative effect of -5 points over the past 12 months.
2. Entry-level work takes the hit. A Stanford working paper using ADP payroll data found a 13% relative employment decline for workers aged 22–25 in the most AI-exposed occupations. The New York Fed put recent-graduate unemployment at 5.7% in Q1 2026, against about 4.2% for all workers. The cause is disputed, because weak hiring and past overhiring also play a role.
3. Projections depend on the time horizon. WEF employers expect 170 million jobs created and 92 million displaced by 2030, a net gain of 78 million. The churn equals 22% of jobs, about one in five. This is a survey projection, not a measurement.

Two cautions apply throughout. Exposure to AI is not the same as job loss, and some employers may be blaming AI for layoffs with other causes (see "Inside the firm"). The report also covers India, where the IT and BPO sectors face the sharpest near-term test.

Scope, method and data cut-off

This report is a synthesis of published sources, not a systematic review of peer-reviewed research. Most sources are institutional reports, firm surveys, think-tank analyses and one Stanford working paper that has not been peer reviewed.

Figures were re-checked on 30 October 2026 against pages found through web search. Several checks relied on press coverage of the original reports, and not every primary document was opened. Where a claim from the first draft could not be confirmed, it was removed or flagged. The scope is global with emphasis on advanced economies, plus a separate section on India.

The older IMF and Goldman figures are exposure estimates made before most of the data above existed, so they are labelled by year wherever they appear.

How large is the net effect?

The sources disagree less than they appear to, because they measure different things over different horizons. Measured outcomes show almost no aggregate effect, a firm survey shows a small negative one, and a 2030 employer projection shows a net gain.

Medium term: net positive, high churn. The WEF Future of Jobs Report 2025 surveyed more than 1,000 employers covering over 14 million workers. They expect 170 million new jobs and 92 million displaced jobs by 2030, a net gain of 78 million. The churn equals 22% of the 1.2 billion formal jobs in the dataset, which is about one in five jobs, not one in four. WEF summary, Engineering News

Near term: modest negative in firm surveys. S&P Global's 2026 report says AI's employment effect has turned modestly negative. Net impact over the past 12 months is -5 points (the share of firms reporting AI-related job losses minus the share reporting gains), with -2 points forecast for 2026. Large firms forecast -13 points, while small firms forecast +3. Only 24% of firms say headcount reduction is a main AI objective, against 64% for process efficiency and 59% for employee productivity. In India the net impact was reported as neutral. S&P Global

Measured outcomes: little aggregate footprint. Yale Budget Lab found no discernible disruption in the US labour market 33 months after ChatGPT's release. Changes in the mix of occupations looked similar to earlier technology waves such as computers and the internet, though Yale cautioned that widespread effects take longer than 33 months to appear. Yale analysis coverage The OECD's June 2026 Economic Outlook reached a similar view at industry level, and found that a shortage of digital skills is slowing AI adoption. OECD coverage

Firm expectations are modest. A McKinsey survey found that two-thirds of respondents saw little or no AI-related change in their total employment last year, while 39% expect AI to reduce headcount in the next year. McKinsey An Atlanta Fed survey found US firms predicting a 1.2% employment reduction over three years. Atlanta Fed

The data do not support a mass-unemployment story today. They support a churn and redistribution story, with a small negative near-term effect in surveys and uncertainty about the 2030 outcome.

Who is exposed?

Exposure is broad but substitution is narrow, and the clearest measured harm falls on early-career workers.

Exposure estimates (older, model-based). The IMF's January 2024 analysis put exposure at about 40% of jobs worldwide, 60% in advanced economies, 40% in emerging markets and 26% in low-income countries. The IMF said about half of exposed jobs in advanced economies could benefit from AI, while for the other half AI could do key tasks and lower labour demand. IMF analysis coverage, Gulf News

Goldman Sachs' March 2023 report estimated that about two-thirds of US and European jobs are exposed to some automation and generative AI could substitute up to a quarter of current work. Within the US it put 7% of workers in jobs where AI could do at least half the tasks, 63% in jobs complemented by AI and 30% unaffected. The 2023 task-automation shares were 46% for office and administrative support and 44% for legal work. These are estimates of task overlap, not forecasts of job loss. Zawya (Reuters), TechBriefly

The entry-level squeeze. Brynjolfsson, Chandar and Chen's August 2025 working paper used ADP payroll data from January 2021 to July 2025. It found a 13% relative employment decline for workers aged 22–25 in the most AI-exposed occupations, even after controlling for firm-level shocks. Employment for older workers in the same occupations and for all ages in less-exposed occupations stayed stable or grew, and the declines were concentrated where AI automates rather than augments work. Benton Institute, The Register

The New York Fed reports recent-graduate (ages 22–27) unemployment at 5.7% in Q1 2026, with the overall rate near 4.2%. Graduates have exceeded the national rate for five straight years. JBizNews The OECD's June 2026 outlook reportedly found unemployment for 20–30-year-olds in AI-exposed roles rising by about 3 points in 2025, though this figure comes from secondary coverage. AI2Work summary

Anthropic CEO Dario Amodei told Axios in May 2025 that AI could eliminate half of entry-level white-collar jobs within one to five years and push unemployment to 10–20%. This is a prediction by the head of an AI company, not a measurement, and the aggregate data to date do not show it. Tom's Hardware

The mechanism proposed in the Stanford paper is that AI automates codifiable early-career tasks such as summarising documents, basic coding and first drafts, while complementing judgment-heavy senior work. The risk is a weaker pipeline of experience, not only temporary youth unemployment. Competing explanations include weak hiring, oversupply of graduates in computing and economic uncertainty, which Yale and Oxford Economics both raise.

India and emerging economies

India's exposure is concentrated in IT services and business process outsourcing (BPO), where the first signs of slower hiring are visible, but the evidence on net job loss is mixed.

Exposure estimates. The IMF's January 2024 analysis put exposure at 40% of jobs in emerging markets and 26% in low-income countries, below the 60% for advanced economies. It said these countries face fewer immediate disruptions but are less ready to capture AI's benefits, which risks widening gaps between countries. Silicon UK S&P Global's 2026 firm survey recorded a neutral net employment impact for India, against -6 for the UK and a positive net impact for Brazil. S&P Global

What the sector data show.

• Tata Consultancy Services, India's largest IT exporter, reported headcount falling from 607,979 to 584,519 in the financial year ending March 2026. The Star
• Nasscom projected about 135,000 net IT hires for FY26, much lower than in earlier boom years. Business Today
• A vendor report says BPO and ITES hiring rose 21.7% year on year in February 2026 and global capability centres (GCCs) may add 425,000–450,000 jobs in 2026. This comes from a commercial blog, so treat it as indicative only. Wisemonk
• A press report on a NITI Aayog study warns that AI could cost India 15–20 lakh (1.5–2 million) IT jobs; I could not open the study itself. Free Press Journal

Predictions versus data. Investor Vinod Khosla said in February 2026 that IT and BPO services could almost completely disappear within five years. This is a forecast, and the hiring and revenue data above do not show anything close to that scale yet. Business Today

Evidence gaps. I did not find reliable measured evidence on AI's effect on India's informal sector or on women's employment. These are the two largest gaps for readers in India, so treat any strong claim on them as unproven until better data appears. Entry-level IT graduates are the group to watch, since the Stanford and OECD findings point to young workers in exposed roles.

Inside the firm: augmentation, automation and AI-washing

Most firms use AI to speed up tasks, and some layoffs blamed on AI appear to have other causes. Both points help explain why high exposure coexists with small aggregate effects.

Augmentation is the main goal. In S&P Global's survey only 24% of firms said headcount reduction was a priority for AI, against 64% for process efficiency and 59% for employee productivity. S&P Global

Agents alongside staff. McKinsey CEO Bob Sternfels said in January 2026 that the firm had about 25,000 AI agents alongside about 40,000 human employees, and expected the two numbers to reach parity by the end of 2026. McKinsey's human headcount had earlier fallen from about 45,000 to about 40,000, with overhiring and economic uncertainty cited as factors. Storyboard18, Fanatical Futurist

Heavy adopters often hire. PwC's 2026 Barometer reports headcount growth of 52% at the most AI-exposed companies against 36% at the least exposed, measured from a 2018 base, and wage growth of 24% against 17%. AI2Work summary of PwC A Ramp and Revelio Labs analysis, reported by the press, found that heavy AI spenders grew headcount about 10% over two years. WebProNews Both come from secondary coverage.

Counter-evidence: AI-washing. Oxford Economics said in January 2026 that firms do not appear to be replacing workers with AI on a significant scale. It noted that the roughly 55,000 US job cuts attributed to AI in the first 11 months of 2025 were only 4.5% of reported cuts, and suspected some firms of presenting layoffs as a technology story instead of admitting past overhiring. Fortune, Allwork.Space A Harvard Business Review survey summary from January 2026 likewise suggests firms are cutting in anticipation of AI's future ability, not its current performance. HBR via Tagteam

Counter-evidence: quality limits. Klarna said in February 2024 that its AI assistant did the work of 700 customer-service agents. In May 2025 its CEO said the cost-cutting push had lowered service quality, and the company began recruiting human agents again, while still planning to shrink overall through attrition. Entrepreneur, Fintech Weekly

A fair reading is that both stories are true at once. Some firms really are cutting roles that AI can do, and some use AI as a label for cuts driven by cost or demand.

Wages and inequality

The AI wage premium is real in job advertisements but easy to misread, because each published figure compares different groups. The table separates them.

Sources: PwC release summaries, AI2Work on PwC, Robert Walters on sector range, Lightcast release, Entrepreneur on the 43% figure.

Three cautions apply. These are advertised wages, not pay received, and employers advertising AI roles may differ in other ways. The 28% and 62% figures use different datasets and methods, so they should not be averaged or compared as a trend. Lightcast also found that 51% of postings requiring AI skills are outside IT and computer science, so the premium is no longer only a tech effect. Staffing Industry Analysts

On inequality, the IMF's 2024 analysis said AI could worsen overall inequality in most scenarios, because higher-income and younger workers may gain more while lower-income and older workers lag. Gulf News The Stanford paper found that adjustment so far has come mainly through employment, not wages. Benton Institute summary If AI skills reinforce existing educational advantages, the gap could widen without deliberate policy.

Scenarios to 2030

These three paths are this report's own illustrative scenarios built from the evidence above, not forecasts from any cited source, and none carries a probability. Their purpose is to show which assumptions drive each outcome and which signals would tell you which path is unfolding.

None of the paths is excluded by current data. Yale and the OECD show no aggregate break yet, but Yale itself said broad effects take longer than 33 months to appear.

Policy and institutional responses

The evidence supports acting on entry-level pathways and skills first, and treats worker voice and tax-base effects as open questions with thinner evidence. Few of these responses have been evaluated against AI specifically.

• Redesign entry pathways. If AI removes routine junior tasks, classroom reskilling cannot replace the learning that early jobs provided. Apprenticeships, structured on-the-job training and junior roles built around AI-assisted work are the logical response to the Stanford and OECD findings. This is an inference from the entry-level evidence; I did not find evaluated programmes.
• Address the skills bottleneck. The OECD's June 2026 outlook said a shortage of digital skills is now a barrier to AI adoption. OECD coverage The IMF has urged comprehensive safety nets and retraining programmes for vulnerable workers. Silicon UK
• Give workers a voice. International bodies such as the ILO and OECD advocate social dialogue when AI is deployed at work. This revision did not re-check their exact wording, and evidence on whether such dialogue improves outcomes is limited.
• Watch the tax base. If AI shifts income from labour to capital, income-tax revenue that funds social protection could shrink. This is a plausible concern raised by fiscal institutions, but I did not verify the specific claims, so treat it as an open question.

The WEF's 78 million net gain is credible only if workers can move into the new roles. That depends on whether training, mobility and support reach the people displaced and the young people entering work.

What this means for you

The practical advice differs by role, but the shared rule is to protect and build the judgment-heavy part of your work. These are suggestions drawn from the evidence above, not guarantees.

Workers

• Check your own task exposure. List what you do in a normal week and mark each task as routine and codifiable (summarising, first drafts, data entry, document review, basic code) or judgment-heavy (decisions, client trust, negotiation, physical presence). The Stanford paper found declines concentrated where AI automates work, not where it assists.
• Use AI to do more of the valuable work. PwC found roles where AI raises the skill content are growing faster than roles it simplifies. Aim to supervise and check AI output in your field.
• Pair AI skills with domain knowledge. AI skills show pay premiums in job ads, including outside IT, but the premiums are advertised figures and not a promise of a raise.

Students and early-career workers

• Build proof of judgment. Portfolios, internships, apprenticeships and real projects show what routine AI tasks do not. Employers hiring juniors increasingly want skills once expected only of senior staff.
• Do not rely on one entry route. Recent-graduate unemployment is above the overall rate, so apply across sectors, including smaller firms and global capability centres, and expect a longer search.
• In India's IT sector, expect slower fresher hiring at the large services firms and keep a second skill, such as data work, a domain such as finance or healthcare, or quality and security roles.
• On "should I learn to code?" Learning to code still builds problem-solving and helps you direct AI tools, but pure junior coding is among the most exposed tasks, so pair it with a domain.

Employers

• Do not cut on a promise. Klarna's reversal and the HBR survey summary show the risk of cutting before AI performs reliably.
• Keep junior roles as learning roles. Redesign them around AI-assisted work, because removing them can weaken your future senior bench.
• Measure outcomes. Track quality, customer results and net headcount, not only the number of agents or tools deployed.

Policymakers

• Track entry-level indicators (see the next section) and fund structured early-career pathways.
• Demand better data. Yale and the OECD both note data limits; employer-reported reasons for layoffs should be audited, given the AI-washing evidence.

Indicators to watch and limitations

Six indicators let readers track which scenario is unfolding without relying on forecasts.

The labour share of national income is a further indicator for the tax-base question, though it moves slowly.

Limitations of this evidence.

• Exposure measures task overlap and does not predict job loss.
• Several key sources are surveys of employers' expectations, which can differ from outcomes.
• The time series is short, about three years since generative AI became widely used.
• The Stanford paper is a working paper, and its authors and others note competing explanations for weak entry-level hiring.
• Employer statements that cuts are due to AI cannot be taken at face value.
• Some figures here come from secondary press coverage, and not every primary document was opened.
• Evidence on India's informal sector and on gender effects is missing.

Key numbers and glossary

The table lists the main figures and what each one actually measures. Wage-premium figures are in the wages section and are not repeated here.

Glossary

• Exposure: the share of an occupation's tasks that AI could perform. It does not mean the job will disappear.
• Augmentation: AI helps a person do tasks faster or better, with the person staying in the loop.
• Automation: AI performs the task in place of a person.
• Complemented job: a job where AI handles some tasks and raises the worker's productivity.
• Churn: jobs created plus jobs displaced, as a share of all jobs. It is larger than the net change.
• Net effect: jobs created minus jobs lost over a stated period.
• Relative decline: a drop compared with a control group, not a fall in absolute numbers.
• AI-washing: attributing layoffs or product claims to AI when other causes, such as overhiring or weak demand, are the main drivers.
• AI agent: software that carries out multi-step tasks with limited human direction.

Frequently asked questions

Will AI cause mass unemployment? The data through mid-2026 do not show it. Yale Budget Lab found no discernible aggregate disruption, and the OECD found no widespread displacement at industry level. The main risks are churn, weaker entry-level hiring and inequality.

Is my job safe? Judge it by tasks, not job title. Roles built on routine, codifiable tasks such as administrative support, legal support, customer service and basic coding are most exposed, and workers aged 22–25 in exposed occupations show the clearest decline. Jobs that rely on judgment, trust and physical presence are less exposed. No job is certain either way.

Which jobs are most at risk? Goldman Sachs' 2023 analysis ranked office and administrative support (46% of tasks) and legal work (44%) among the most exposed. Customer service and routine programming also appear repeatedly in recent studies.

Which jobs are growing because of AI? The WEF lists AI and machine-learning specialists, big-data specialists and fintech engineers among the fastest-growing roles, alongside care and frontline jobs driven by demographics. PwC finds jobs requiring AI skills grew much faster than the overall market.

Does AI create more jobs than it destroys? The WEF's employer survey projects a net gain of 78 million by 2030. S&P Global's firm survey shows a small net negative effect now. The answer depends on the time horizon and on how fast workers and institutions adapt.

What is the AI skills premium? PwC finds a 62% advertised pay premium for roles requiring AI skills, and Lightcast finds 28% for postings listing one AI skill. They are different measures from different datasets, and both are advertised pay.

Should I learn to code? Coding still teaches problem-solving and helps you direct AI tools. But basic junior coding is among the most exposed tasks, so combine it with domain knowledge, such as finance or healthcare, and with judgment skills.

How do I future-proof my career? Keep a second skill, use AI tools to raise your output, build evidence of judgment through projects and experience, and watch the indicators listed above. No strategy removes the risk, but these reduce your dependence on routine tasks.

Is India's IT sector finished? The data do not support that. Large firms such as TCS reduced headcount in FY26 and Nasscom projected lower hiring, but other indicators, such as GCC hiring, are growing. Forecasts of the sector disappearing within five years are predictions, not measurements.

What should policymakers do? Prioritise structured entry-level pathways, skills and safety nets, better data on why employers cut jobs, and attention to the tax base if income shifts from labour to capital.

Sources and corrections

Sources were retrieved through web search on 30 October 2026. Many are press coverage of the original reports, so the primary documents should be consulted before citing a figure in formal work.

• S&P Global, AI impact on employment 2026
• Yale Budget Lab analysis, via diginomica
• OECD June 2026 outlook, via Outsource Accelerator
• WEF Future of Jobs 2025, via eTradeForAll
• IMF exposure analysis, via Gulf News
• Goldman Sachs 2023, via Zawya
• Stanford "Canaries" paper, via The Register
• New York Fed recent-graduate data, via JBizNews
• PwC 2026 Barometer, via Ground News
• Lightcast Beyond the Buzz
• Oxford Economics on AI-washing, via Fortune
• McKinsey survey on AI and employment
• Atlanta Fed on firm AI spending and headcount
• Klarna reversal, via Entrepreneur
• Amodei interview, via Tom's Hardware
• India IT: The Star on TCS headcount