AI Is Changing Tasks, Not Erasing Jobs: What the Global Employment Data Actually Shows for Young Workers
Most writing about AI and jobs makes one of two claims, and both are wrong in the same way. Either AI is about to eliminate entire professions, or AI changes nothing and this is another hype cycle. The published data supports neither. What it supports is a narrower and more useful statement:
AI is currently reorganising tasks inside occupations much faster than it is eliminating occupations, and the people most exposed to that reorganisation are the ones trying to enter the labour market, not the ones already in it.
This post works through what the primary sources actually measured, including where they contradict each other. It is written for students choosing a path, researchers who need the citations, and professionals deciding what to learn next.
1. The headline numbers, and what each one is
Before any argument, it helps to separate observed data from employer expectations from modelled forecasts, because these get quoted interchangeably and they are not the same kind of claim.
| Figure | What it says | Evidence type | Source |
|---|---|---|---|
| 5.0% | Global unemployment rate, a historic low | Observed | ILO WESO Trends 2025 |
| 12.6% | Global youth unemployment rate | Observed | ILO WESO Trends 2025 |
| 25% | Share of global workers in occupations with some GenAI exposure | Modelled task estimate | ILO, Generative AI and Jobs, 2025 |
| 3.3% | Share of global employment in the highest AI task-exposure band | Modelled task estimate | ILO, Generative AI and Jobs, 2025 |
| 170M / 92M | Roles created vs displaced globally by 2030 | Forecast and employer expectation | WEF Future of Jobs 2025 |
| 39% | Share of workers' core skills employers expect to be disrupted by 2030 | Employer expectation | WEF Future of Jobs 2025 |
| 43% | Share of OECD bachelor's entrants who graduate on time | Observed | OECD Education at a Glance 2025 |
Notice how much of the alarming material is in the "expectation" and "forecast" columns, and how much of the observed data is comparatively calm. That asymmetry is the real story.
2. Exposure is not displacement
The ILO mapped generative AI capabilities against roughly 30,000 occupational tasks. The result:
Why the gap between 25 percent and 3.3 percent is so wide: almost every real occupation is a composite of physical, contextual, interpersonal and cognitive tasks. Generative AI is strong on one of those four. An occupation becomes fully substitutable only when nearly all of its task bundle falls into that one category, which is rare.
This is why the WEF's net projection is positive. Weighing 170 million roles created against 92 million displaced by 2030 gives a net gain of about 78 million.
That last line matters and is almost always omitted. The occupations with the highest absolute projected growth are agricultural workers, driven by climate adaptation, and care and nursing professionals, driven by population ageing. Both are among the least exposed to generative AI. The fastest-declining occupations are data entry clerks, administrative assistants and payroll clerks: roles whose task bundles are almost entirely routine text and data processing.
| Occupation category | Growth trajectory | AI / automation exposure | Primary driver |
|---|---|---|---|
| AI and machine learning specialists | High growth, percentage terms | Low (they build the systems) | Technology adoption |
| Agricultural professionals | High growth, absolute terms | Low (physical, unstructured) | Climate adaptation, supply chains |
| Care and nursing professionals | High growth, absolute terms | Low (requires human presence) | Ageing demographics |
| Clerical and administrative | High decline | High (routine text and data) | GenAI language capability |
| Media and content design | Transforming, not declining | Medium to high | Task augmentation |
Source: WEF Future of Jobs 2025; ILO 2025.
3. The entry-level question, with the counter-evidence
The most repeated claim right now is that AI is destroying entry-level work by absorbing exactly the junior tasks that graduates used to be hired for: summarising documents, basic coding, drafting copy. The mechanism is plausible. If organisations restructure around AI agents for routine coordination, the traditional corporate apprenticeship breaks: the market demands experience while shrinking the roles through which experience was acquired.
That is a real structural risk and it deserves to be taken seriously. But the best real-time data available does not yet confirm it.
LinkedIn's Economic Graph analysis (January 2026, covering 1.3 billion members across the US, UK, India, France and Germany) found:
- Macroeconomic conditions, not AI exposure, explain most US hiring variation since 2022. Real interest rate movements track the hiring curve closely. The LinkedIn Hiring Rate sat about 23 percent below pre-pandemic levels in 2025.
- Entry-level hiring has held up slightly better than experienced hiring, not worse, averaging about 3 percentage points above manager and senior-level hiring over the last two years.
- Software engineering entry-level hiring tracks overall software engineering hiring. It softened in late 2025 (down about 5 percent year on year in December) but moved with the broader cycle. LinkedIn's own conclusion: it is too early to attribute specific shifts to AI.
- Hiring shows no significant difference by AI exposure. In fact "augmented" occupations, which mix AI-replicable tasks with human-centric skills, such as data analyst and web developer, outperformed overall hiring, closing 2025 at +2.8 percent year on year.
So the honest position is: the graduate labour market is genuinely difficult, and the evidence currently attributes most of that difficulty to interest rates and post-pandemic normalisation rather than to AI. The experience gap is a moderate-confidence risk about where things are heading, not an observed cause of where they are.
Anyone telling you otherwise with certainty is ahead of the data.
4. Youth unemployment is the structural problem
This is where the numbers are unambiguous and grim.
- Global youth unemployment: 12.6 percent, roughly triple the adult average.
- Youth NEET (not in employment, education or training): 28.2 percent for young women against 13.1 percent for young men, driven largely by unpaid care responsibilities.
- Global labour force participation: 61 percent, falling in high-income countries, rising in middle-income ones.
And the gap is not evenly distributed. In India, youth unemployment runs at about four times the non-youth rate, well above the global pattern. Roughly 11 million of 63 million Indian graduates aged 20 to 29 were unemployed as of 2023, and tracking young graduate men for a year from the point they report unemployment, only about 7 percent find permanent salaried work. (State of Working India 2026; I go into the Indian system in detail in the companion post on Indian universities.)
The structural explanation is demographic, and it splits the world in two:
GLOBAL NORTH GLOBAL SOUTH
(North America, EU, East Asia) (Sub-Saharan Africa, South Asia)
------------------------------ --------------------------------
Ageing workforce Youth bulge
Labour SHORTAGE Labour ABSORPTION failure
Retain older workers Informal work as default
Automate aggressively Brain drain accelerates
Need: adult reskilling Need: scalable foundational
and vocational education
These are not the same problem and they do not have the same solution. A policy imported from one side to the other will fail.
5. Cross-border work is where the leverage is, and the risk
If you are a young worker in a youth-bulge economy, the structural implication is that the demand is somewhere else. The data on how that gap gets bridged is striking.
Talent concentration. High-income countries hold 53 percent of global ICT specialists and 70 percent of AI job postings, while upper-middle-income countries hold 29 percent of ICT talent with 16 percent growth in AI postings between 2021 and 2024. In some low and lower-middle income markets, talent outflows run 3 to 4 times inflows. (World Bank, Digital Progress and Trends 2025)
Mobility premium. AI engineering professionals are 8 times more likely to move across borders than the average LinkedIn member. Portable skills travel.
Demand is growing fastest outside the traditional hubs. Job postings requiring AI engineering skills grew +54 percent year on year in India and +37 percent in the UK in 2025, while demand levelled off in France, Germany and the US.
Infrastructure is a job category now. The global data centre workforce has doubled since 2017, with postings up 23 percent year on year in 2025 and over 600,000 net new global jobs created in a year. The US holds about 40 percent of that workforce; Germany is the fastest-growing hub.
The World Bank frames regional capacity to participate as the 4Cs:
CONNECTIVITY -> infrastructure to reach the work
COMPUTE -> data centres to do the work
CONTEXT -> local data to make the work relevant
COMPETENCY -> skills to perform the work
A region missing any one of the four cannot capture value from the AI economy regardless of how many graduates it produces. India, notably, is building compute and has competency growth, which is a better starting position than most middle-income peers.
The outbound student counterpoint
Cross-border mobility also runs the other way, as capital. Indian outbound student mobility grew from 6.84 lakh in 2016 to 13.35 lakh in 2024, while inbound international students stayed roughly flat (45,424 in 2016 to 46,878 in 2022). At the 2021 peak, 24 Indian students went abroad for every 1 who came in.
Estimates of what that costs vary widely by methodology and should be treated as estimates, not measurements. One widely cited projection puts total expenditure by Indian students on overseas education at roughly USD 70 billion by 2025, which the same working paper notes is on the order of 10 times the Government of India's annual higher education budget. Treat the exact figure with caution; the direction is not in doubt.
The strategic read: remote and cross-border work lets a young worker in a youth-bulge economy access high-income demand without the forex outflow, visa risk and migration cost of physically relocating. That is a genuinely new option, and it is almost entirely gated on demonstrable, verifiable skill rather than on local credential prestige.
6. Which skills the evidence actually supports
Employers expect 39 percent of current core skills to be disrupted by 2030. The fastest-growing demanded skills are a pairing, not a single category.
| Rank | Skill | Category |
|---|---|---|
| 1 | AI and big data | Technical |
| 2 | Cybersecurity | Technical |
| 3 | Technological literacy | Technical |
| 4 | Creative thinking | Human |
| 5 | Resilience, flexibility, agility | Human |
Source: WEF Future of Jobs 2025 (employer expectation).
Within the technical side, LinkedIn's 2025 data gives a sharper picture of what is actually moving:
| Fastest-growing AI engineering skills | Fastest-growing AI literacy skills | Top occupations adding AI literacy |
|---|---|---|
| AI agents | AI prompting | Software engineer |
| AI strategy | Microsoft Copilot Studio | IT architect |
| LLMOps | Prompt engineering | IT consultant |
| AI productivity | Microsoft Copilot | Product manager |
| LangChain | GitHub Copilot | Graphic designer |
Two things stand out. AI agents was the fastest-growing AI engineering skill of 2025, which suggests organisations have moved from experimentation to workflow redesign. And AI literacy has spread well past engineering: product managers and graphic designers now rank in the top five occupations adding these skills.
The corporate picture from Mercer's Global Talent Trends 2026 fills in the demand side, and it contains the most uncomfortable number in this entire post:
- 63 percent of executives say redesigning work to incorporate AI is the people initiative that will drive the greatest return, but only 32 percent believe their workforce can optimally combine human and machine capability.
- 98 percent of executives plan organisational design changes in the next two years, and 99 percent expect AI to lead to at least some headcount reduction.
- 54 percent of the C-suite name talent scarcity as the top macro driver of people plans; 59 percent of HR leaders say difficulty attracting talent with vital digital skills is their top challenge.
- Employee thriving fell from 66 percent in 2024 to 44 percent in 2026.
- 35 percent of employees would consider leaving if they felt disadvantaged by unequal access to AI tools or training.
Read those together: organisations expect to cut headcount and cannot find the skills they need and their existing people are depleted. That combination is not "AI replaces workers." It is a capability mismatch at scale, which is a much better situation for a well-prepared entrant than a pure contraction would be.
7. What a degree is worth now
The claim that "degrees are becoming worthless" is not supported.
Tertiary education still commands a substantial earnings premium and remains the primary insurance against unemployment across OECD markets. What has changed is the function of the credential. It has moved from a proxy for occupational competence to a baseline signal: it says perseverance, broad cognitive ability and social capital. It no longer says task-specific competence. That has to come from somewhere else, which is why employers adopting skills-first frameworks are doing so to augment degrees, not to replace them.
The weak point is completion, not value.
And access remains stratified by background: 70 percent of young adults with at least one tertiary-educated parent attain a tertiary qualification, against 26 percent of those without. Expansion of access has not dissolved inheritance.
The emerging model the evidence supports is modular:
- Stackable microcredentials, because specific technical skills now evolve faster than multi-year curricula can be accredited.
- Work-integrated learning, which is the only direct answer to the experience gap.
- Competency-based assessment, moving away from time-in-seat.
- Continuous lifelong learning, which the WEF identifies as simultaneously one of the weakest current workforce competencies and one of the most necessary.
The catch, which the OECD is blunt about: adult learning participation is highly unequal and heavily favours people who already hold tertiary credentials. The reskilling economy currently compounds advantage rather than redistributing it.
8. What this means in practice
For students: the degree is still worth finishing, and finishing matters more than the data suggests you have been told (43 percent on-time completion is the real risk, not AI). Pair it with evidence an employer can inspect. The occupations with the best combination of growth and low substitution risk are either deeply human (care, health, skilled trades) or build the systems (AI engineering, data infrastructure, cybersecurity). The middle, routine cognitive work with low human-contact, is where the decline is concentrated.
For researchers: the open questions that the current datasets cannot answer are, in my view: where graduates will acquire practical experience if junior tasks are absorbed; how institutions should integrate verifiable capability portfolios alongside transcripts; and how to authentically assess "human-centric" skills within a standardised framework. All three are measurement problems before they are policy problems.
For professionals: the pairing is the point. AI literacy alone is becoming table stakes (it is spreading into product and design functions, not just engineering). The scarce combination is domain depth plus the ability to restructure a workflow around AI, which is exactly the capability 63 percent of executives say they want and only 32 percent believe they have.
What the data does not tell us
- Productivity effects are unresolved. It remains genuinely uncertain whether AI implementations will translate into measurable macroeconomic productivity gains. This is the Solow paradox in new clothes.
- The informal economy is largely invisible here. These datasets skew heavily toward formal, digitally visible employment. For countries where the majority of work is informal, the AI impact is poorly captured.
- Long-term wage effects are speculative. Whether AI polarises wages or raises the productivity floor for lower-skilled workers is not settled.
- LinkedIn data reflects LinkedIn members, which skews toward white-collar, English-language and formally employed populations.
- Employer surveys measure intent, not outcome. The 170M/92M figures and the 39 percent skills-disruption figure are what employers expect, which has historically been a noisy predictor.
References
- International Labour Organization, World Employment and Social Outlook: Trends 2025. https://www.ilo.org/publications/flagship-reports/world-employment-and-social-outlook-trends-2025
- International Labour Organization, Generative AI and Jobs: A Refined Global Index of Occupational Exposure, 2025. https://www.ilo.org/
- World Economic Forum, Future of Jobs Report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
- OECD, Education at a Glance 2025. https://www.oecd.org/education/education-at-a-glance/
- World Bank, Digital Progress and Trends Report 2025. https://www.worldbank.org/
- LinkedIn Economic Graph, AI Labor Market Update, January 2026. https://economicgraph.linkedin.com/
- Mercer / Marsh, Global Talent Trends 2026. https://www.mercer.com/insights/people-strategy/future-of-work/global-talent-trends/
- State of Working India 2026, Centre for Sustainable Employment, Azim Premji University. https://azimpremjiuniversity.edu.in/centre-for-the-study-of-the-indian-economy-csie/cse
- Internationalisation of Higher Education in India: A Temporal Overview, working paper on international student mobility.
- UNESCO, guidance on generative AI in education and research. https://www.unesco.org/en/digital-education/artificial-intelligence
Figures are reproduced from the sources named in each caption. If you think I have read one of them wrong, tell me and I will correct it.