By Aarav Beniwal
Recent evidence suggests that AI’s impact on employment is convoluted and varied across regions and industries. Surveys and labor-data analyses paint a nuanced picture, job counts are generally stable or rising even as tasks and skills shift. Global analysis finds that almost every type of AI-exposed job saw growth from 2019-24. That would mean, AI often augments work rather than outright replaces workers. Policymakers and companies now debate how quickly this will continue, how many roles will be redesigned or displaced, and what training or support may be needed.
Studies show the fastest AI adoption in North America and developed Asia, with developing economies often leading in new uptake. Within industries, knowledge-intensive and white-collar sectors including tech, finance, professional services, and healthcare inter alia face the most AI disruption, while manual and trade-heavy sectors see relatively little so far. Younger and entry-level workers appear more exposed to AI-driven change than older, senior workers. Importantly, AI tends to automate tasks, often routine or administrative duties, not entire occupations or at least yet. A McKinsey report finds that although technology could in theory automate ~57% of US work-hours , much of that potential remains unrealised or too costly, meaning only a minority of tasks are fully automatable now. Empirical data show partial task replacement and job augmentation dominating current AI effects.
Looking ahead, most forecasts still expect net job creation alongside transformation. The World Economic Forum’s Future of Jobs report projects a net 78 million new roles globally by 2030 even as 22% of current jobs are reshaped. At the same time, 85% of employers plan substantial retraining for their workforces, reflecting the need for policy focus on upskilling. In light of these trends, experts are debating how best to support workers through the AI transition via education, training, social safety nets and ensure the economic gains of AI are broadly shared.
Regional and Demographic Variations
In the US and Canada, AI adoption is already widespread in business. A McKinsey report found that 90% of US companies use some AI, and AI-fluency demand grew sevenfold in two years. Tech companies and high-skill industries lead AI use. Labor-market indicators have remained tight even as AI tools proliferate. A Stanford/ADP analysis shows overall employment in AI-exposed occupations like software development and customer service remains stable or growing for mid-career workers, even while entry-level hires have sharply declined. Specifically, the ADP data chart below shows that after late 2022, the advent of generative AI, 22–25 year old software developers (blue line) and customer service reps saw a drop in headcount, whereas older cohorts saw flat or rising employment. Health-aide jobs with low AI exposure continued rising for all age groups, illustrating the targeted nature of the effect
Only about one-third of Americans trust that companies will use AI responsibly, and many workers worry about job impact. Globally, only 57% of workers expect AI to change their current jobs, and 36% expect it to replace them. In the US specifically, public nervousness about AI is high, Anglophone countries report the most concern, while Japan and South Korea report the least. On the other hand, productivity growth has been strong in AI-intensive US industries. One analysis found that AI-exposed sectors in the US saw productivity growth quadruple from 7% to 27% since 2022 , and U.S. workers with AI skills earned a 56% wage premium in 2024.
Sentiment and adoption patterns are even more varied in the Indo-Pacific. In many countries, workers are more optimistic or at least more prepared for AI. A Deloitte survey of ~12,000 people across 13 Indo-Pacific economies found that developing economies are ahead of developed ones in GenAI adoption (adoption rates 30% higher). In Southeast Asia, over 90% of students and 72% of workers reported having used generative AI, and 65% believed their tasks would be automated or augmented within five years. Deloitte also estimates that GenAI will affect 11 billion work-hours per week across APAC. Similarly, 60%–77% of APAC employees think AI will enhance their economy’s role in the world.
Yet many Asian workers also fear disruption. An Ipsos survey found the highest job-loss anxiety in emerging Asian markets, 69% of Thai workers and 62% of Malaysians and Indonesians expect AI to replace their jobs, compared to only 32% of Americans saying they trust employers to use AI safely. Not surprisingly, younger, tech-fluent “Gen AI” workers are driving Asia’s rapid adoption, but only about half believe their managers fully understand what they’re doing .
Meanwhile, Europe’s data are mixed. A broad global survey found lower excitement and higher skepticism about AI in Western Europe, similar to the US pattern. By contrast, an IMF study modelling advanced economies predicts net job growth from AI in the long term with sectoral shifts and urges immediate action on training and social insurance.
In Latin America and Africa, concrete data are sparse, but experts note that limited infrastructure and education will blunt AI’s direct impact. The World Bank’s analysis of 25 low- and middle-income countries found only 12% of workers in the poorest nations are “exposed” to AI vs 15% in lower-middle. Therefore, while Asia and North America lead the AI shift, African and low-income countries likely face more augmentative than disruptive effects for now.
Industry and Occupation Effects
The sectoral footprint of AI is highly uneven. Consistently, information and communications technology (ICT) and other white-collar fields show the largest AI hiring and investment. PwC states that in the US the share of ICT job postings requiring AI skills jumped from only 1.4% in 2012 to 9.5% in 2024, the fastest growth of any sector. By contrast, sectors like healthcare, social care, and construction had almost no AI skills demand, each <1% of postings, in 2024. In other words, tech and knowledge industries, especially finance, professional services and high-end manufacturing like aerospace are racing ahead in AI, while manual and routine industries lag far behind.
Many studies emphasise that cognitive, information-based tasks are most exposed. Generative AI excels at writing code, reports, and handling data, precisely the duties of software engineers, data analysts, financial advisors, and legal researchers. Brookings, a Washington D.C. based think tank, reports that “AI is especially well suited to the cognitive tasks of white-collar knowledge work ,” whereas it “is currently not equipped to handle the manual work” done by manufacturing, construction, skilled trades, or many service jobs. Advancement in Robotics and associated technologies may change that. Accordingly, AI exposure tends to track with wages, better-paid, better-educated occupations face higher AI exposure.
Even within exposed fields, task-level differences matter. McKinsey analysis finds that within broad occupations, roughly two-thirds of work is “nonphysical” such as information-processing or communication, and of those hours, about one-third require social/emotional skills beyond AI’s reach. Many of the remaining tasks involve abstract reasoning or language, well-suited to AI, which together account for roughly 40% of US wages. This means roles heavy on rote writing, basic research, or data aggregation see more automation, while jobs centered on collaboration, creativity or high-touch service remain largely intact. While chatbots can answer first-level customer questions, they cannot replace the judgment of a doctor or the empathy of a nurse – at least not yet.
On net, industries oriented around information have already seen AI-driven productivity surges. One report finds that since 2022, productivity growth in highly AI-exposed industries has accelerated four-fold to 27% annual growth, outpacing their less-automated peers. Another finds ICT firms adding the equivalent of 9% of global revenue from GenAI, with banking, pharma and education up to 5% each. By contrast, sectors relying on physical labour show little such change . This suggests that the “AI era” may mirror past tech waves, it creates new industries and augments innovation in knowledge sectors, while older routine industries evolve more gradually.
Hierarchy and Worker Level
An emerging finding is that AI’s disruption is most profound at the bottom of the career ladder. Across occupations, entry-level or junior positions with simpler, more routine tasks are most easily encroached upon by AI. The ADP/ Stanford study clearly illustrates this, in both software development and customer service, young workers between ages 22-25 saw a sharp decline in employment after Oct 2022, whereas more experienced cohorts of ages 31-50+ saw flat or rising employment. In contrast, an occupation with low AI exposure grew across all age groups. This pattern suggests companies may be using AI to accelerate tasks traditionally assigned to trainees or to reduce hiring for entry-level needs, while still valuing skilled, experienced staff.
Such age effects echo other findings. McKinsey’s global survey notes that workforce skill needs are shifting, with employers reskilling millions but only a few anticipating massive layoffs. Stanford researchers comment that “over the worker lifecycle, as tasks become more complex and harder to automate, AI tools tend to augment rather than replace work”. AI can handle simpler sub-tasks such as code snippets or call scripts which compress entry roles, but senior staff use AI as a tool to boost productivity rather than being replaced.
At higher levels, the effect is generally one of more augmentation. Executives and professionals increasingly use AI for decision support, forecasting, data analysis, and content drafting, making their work more productive. For example, 28% of surveyed corporate leaders say generative AI is already on their board’s agenda. However, such leaders mainly see AI as a competitive advantage rather than a headcount reducer; McKinsey finds that while AI usage prompts workforce shifts, it also spurs heavy investment in upskilling.
In middle management and technical roles, AI reshuffles tasks more than jobs. Many middle-tier occupations comprise both automatable components such as data entry or scheduling and non-automatable components. With AI, mid-level workers can offload routine work using AI to generate a first draft of a report and focus on oversight and complex problem-solving. Some tasks formerly done by managers are now partially handled by AI, but the managerial roles themselves remain.
Across the hierarchy, the emerging consensus is that AI changes “how” people work more than “whether” they have work. The World Bank notes that “exposure” does not mean job loss, it can mean task automation or productivity gains. McKinsey similarly projects that most occupations will be reshaped but not eradicated, with “work increasingly centered on collaboration between humans and AI”. Current data show entry-level roles bearing the brunt of AI’s advance, while mid-and upper-level roles evolve with AI augmentation.
Tasks and Complexity: Partial vs Complete Automation
It is critical to distinguish between automating tasks and automating whole jobs. Nearly all studies emphasize that partial automation (AI taking over some tasks) far outweighs full job replacement at present. For example, a new MIT analysis notes that while about 50% of work tasks might be technically automatable by current AI models , only about 23% of worker effort exposed to AI is actually cost-effective for firms to automate now . High upfront costs and integration challenges mean many “exposed” tasks will instead be augmented.
This resonates with labor data. The PwC Global AI Jobs Barometer reports that even roles with the highest automation potential grew by 38% in headcount from 2019–2024 . In other words, companies are hiring more people in AIsusceptible roles – often to complement AI – not slashing those jobs. Similarly, the ADP/Stanford team finds that junior AI-exposed roles declined 2022–25 (as above), but overall employment in those categories (including older workers) held steady or rose, underscoring a rebalancing of tasks rather than emptying of positions .
McKinsey’s modelling aligns with this: today’s AI can technically handle ~57% of US work-hours , but actual adoption is limited by cost, policy, and the time it takes workplaces to adapt . Indeed, only certain subsets of tasks (e.g. routine data processing or language queries) are economically worth automating now . As McKinsey notes, 70% of skills used in automatable jobs are also used in non-automatable jobs – meaning workers simply shift their skill usage.
In effect, AI is automating “pieces” of work. A customer service representative might see an AI chatbot handle firstlevel queries, but complex customer issues still need a human. A junior analyst might get an AI-generated first draft of an analysis to refine, rather than writing it from scratch. Many tasks – especially those requiring judgment, empathy, or creativity – remain safely in human hands. For example, Brookings highlights that AI can do “the rote cognitive tasks” of customer service or clerical work, but not “in-person service” or the fine motor and situational awareness needed in manufacturing or construction .
Examples of Partial Impact
- Legal profession: A U.S. report notes the “legal profession had the second highest (AI) exposure, with an estimated
44% of tasks susceptible to automation”. This refers mainly to tasks like document review or contract drafting, not
the entire lawyer role. Law firms increasingly use AI for research and drafting, freeing lawyers for strategy and
advocacy. - Programming jobs: AI coding assistants can automate boilerplate coding and bug fixes, but software jobs entail
design, system architecture, and client interaction. The ADP data chart shows early-career developers saw declines,
but mid-career and senior developers saw rising employment, implying firms may hire fewer juniors but more
experienced devs who can effectively use AI tools. - Creative fields: Generative AI can produce first-drafts of text, images, or ideas, yet human editors and creatives are
needed to refine, oversee, and ensure quality. In sectors like marketing and design, use of AI has skyrocketed, but
it’s reshaping roles (marketer-with-AI) more than eliminating them.
Overall, empirical data do not support narratives of wholesale robotic job death. Even in highly automatable roles, total headcount is stable or growing, wages in AI-exposed jobs are rising, and demand for AI skills is surging. This aligns with historical patterns, past innovations often killed tasks faster than jobs, giving labor markets time to adjust. As one labor analyst quipped, “AI will make some skills less important, but we’ve seen technology change tasks for decades without mass unemployment”.
Empirical Data Highlights
Across North America and Europe, job postings requiring AI skills have exploded. Gloat reports that in the US, occupations explicitly demanding AI fluency grew from 1 million workers in 2023 to over 7 million in 2025, marking a sevenfold increase and “the fastest-growing skill category”. By late 2024, roughly 16,000 US jobs per month mentioned AI, and roles requiring generative-AI expertise have quadrupled in two years.
Workers with AI skills command a huge premium. PwC’s analysis of job ads finds a 56% wage premium for AI-skilled positions in 2024 up from 25% in 2023. McKinsey and PwC both report that companies using AI see surging output as AI-intensive firms are pulling away with much higher productivity growth.
Charting AI demand by sector reveals stark contrasts. In the US, PwC shows ICT roles in tech and communications now lead the way, nearly 10% of postings, while education, healthcare, finance and manufacturing trail behind. Globally, McKinsey predicts that tech, banking, pharma, and education could see 4-9% boosts in value from GenAI, while autos and aerospace might see only 1-2%, mirroring the pattern that knowledge work is AI’s sweet spot.
Surveys and models indicate most jobs face only partial change. Stanford’s data science team found that younger workers in exposed roles lost ground, but older cohorts filled different roles or took on new tasks. McKinsey’s skillindex suggests 60-70% of hours could be automated by 2030, in theory, but only 20-30% might actually shift based on adoption rates. Indeed, in modelling one sector, the IMF finds a 20-30% drop in employment offset by a 15% rise in wages as productivity increases, and long-run unemployment even falling as other sectors expand. The common thread, jobs evolve with AI, rather than vanish.
Complementary skills are surging. Beyond core AI expertise, firms now prize data analysis, quality control, teaching, and process optimisation skills. Most workers will not need to become AI experts, but they will need to collaborate with AI in ways such as prompt-writing or oversight. Gartner predicts 80% of software engineers must upskill by 2027 for AI, and the World Economic Forum estimates 60% of all workers will need retraining by 2030. This underpins calls for major reskilling efforts.
The Future: Trends and Projections
Most forecasts agree that AI will reshape jobs extensively, but yield net growth or transformation rather than pure shrinkage. The scenario with generative AI is still evolving, but some emerging projections include the WEF’s 2025 Future of Jobs report, which projects 78 million new jobs globally by 2030 even after accounting for shifts and automation. It also finds 85% of companies are already training staff or plan to, indicating anticipation of new work. In the US, McKinsey’s scenario analysis suggests that preparing workforces could unlock $2.9 trillion in economic value by 2030, implying many productivity-driven new opportunities.
The consensus is that even powerful AI takes time to enter workplaces at scale. Historical analogies suggest multiyear lags. The U.S. Bureau of Labor Statistics underscores that historically, “job displacement has occurred in the past, but it tends to take longer than technologists typically expect”. Early indicators support that view, aside from certain tech roles, the main effect so far has been shifting job content, not mass layoffs.
Even as total jobs grow, many roles will change fundamentally. McKinsey finds that roughly 70% of skills are used in both automatable and non-automatable tasks, meaning most workers will need to reallocate time from tasks that AI can do to tasks that only humans can. Routine activities may become obsolete, while managing AI, empathic interaction, and creative problem-solving become more important. Training and workforce planning assume most workers adapt to new tasks.
Long-range economic impact estimates vary widely. Some optimists including Korinek posit AI could double output in 15 years; others such as Acemoglu project much more modest GDP gains of 1-2% over a decade. Outcomes will depend on technology rollout, regulation, and how well people and institutions adjust. Crucially, most analysts stress that AI’s rollout is not preordained; policy and business choices will heavily influence whether gains are broad or concentrated.
Policy Implications and Debates
In light of these findings, what policy questions arise? Experts generally call for proactive measures, focusing on equipping workers for change rather than resisting AI outright. Primarily, with AI shifting skill demands rapidly, lifelong learning and retraining programs are critical. Surveys show 85% of employers intend to retrain or upskill existing staff. Government and corporate training initiatives, for example, expanding STEM and AI literacy from K-12 through adult education, are widely recommended. As the McKinsey report notes, “employers are already adjusting: demand for AI fluency has jumped nearly sevenfold in two years”, so “helping today’s workers build those skills is a must”.
Even if net job losses are small, displaced workers and communities will feel real disruption. The IMF and labor economists argue for robust unemployment insurance, job-search support, and sectoral mobility programs. If AI rapidly shrinks jobs in one sector, workers need time and support to retrain for others. Economic modelling shows that welldesigned UI, perhaps temporarily higher benefits or extended coverage during transition, can cushion income shocks without discouraging job search. These debates echo past ones about how to share transition costs fairly.
Since most evidence suggests AI boosts productivity, some propose policies to accelerate beneficial use such as R&D tax credits for AI tools in healthcare and education. Others warn that laissez-faire can lead to inequality, the Bernie Sander’s staff, from the office of the long serving U.S. Senator, report painted a dire view of 100 million US jobs at risk, though most economists consider that an overestimate. A more balanced proposal is to align incentives so businesses invest in human capital alongside AI capital, subsidising worker training or requiring “high-road” practices if AI is used to downsize.
Longer-term discussions include ideas like a “robot tax” or universal basic income. While these are politically contentious, they arise from concerns that AI could concentrate wealth among tech owners. The World Bank and OECD have suggested exploring revenue options to fund retraining and social spending. Any such policy would need careful design to avoid stifling innovation or hurting growth.
Broader regulation won’t be discussed here in detail, but labor policy will intersect with these. For instance, protecting gig workers who lose tasks to AI, or ensuring that automated hiring tools do not embed bias, are emerging issues. Governments are beginning to include labor-impact assessments in their AI strategies.
The evolving picture is that AI will transform jobs significantly but not end employment – at least as of now. Companies and governments should plan for an era of rapid task change, training workers in AI collaboration, refreshing education curricula, and upgrading social programs. The exact magnitude of job losses or gains remains uncertain, but evidence so far points to shifting roles rather than mass unemployment. Policy debates thus center on preparing people through reskilling, mobility, and income support so that the economy can reap AI’s productivity gains while minimising pain for workers.
Looking ahead, data-driven forecasts see AI continuing to spread, with acceleration expected in the next few years. Organisations that adopt AI tools effectively and prepare their workforce often gain competitive advantage. Workers who adapt, combining technical AI fluency with complementary human skills, are likely to thrive. But without active intervention, there is a risk of widening inequalities between regions, sectors, and skill levels. These findings argue for urgent but balanced policy action, investing in skills, updating education, and strengthening the social safety net, rather than trying to halt innovation.