Context

The world of work is undergoing a profound transformation, brought about by digitalisation, automation, robotisation, algorithmic management and – most recently – the introduction of artificial intelligence (AI). These technologies change the way work is done and valued, and the skills and resources needed, and affect employment and working conditions. The advent of generative AI (GenAI) has further accelerated this process.

AI affects the task contents, workflows and job structures. On the one hand, GenAI can enhance human skills, substitute for knowledge among newer or average-performing workers (cognitive extension), and empower less specialised workers to perform some expert tasks (democratisation of expertise). On the other, it can degrade human skills, as offloading work to AI can diminish critical thinking, metacognitive awareness, and creativity. As the current phase of AI adoption still largely focuses on AI tools that assist work without reorganising it, the available evidence only captures early signals of impact. It is expected that agentic AI systems (capable of planning, coordinating and executing multi-step workflows) will be able to accelerate and amplify AI's impact on labour markets significantly.

Estimating impact

AI's impact on jobs is estimated from their degree of exposure to the technology. Depending on the unit of analysis chosen – a whole occupation, the tasks within an occupation, or the abilities necessary to pursue an occupation – researchers adopt either a task-based or an ability-based approach. Most of the current studies parse jobs into tasks and then estimate to what degree these can be automated, or performed by AI more efficiently than by humans.

AI-exposed jobs comprise multiple tasks or actions for which AI can be used, as is the case, for instance , for data entry workers or financial analysts. However, occupations with similar AI-exposure may have different AI complementarity, meaning the degree to which AI can complement rather than substitute a human. Consequently, analyses distinguish between augmentable jobs, where AI enhances or supports human judgement and expertise (e.g. judges, surgeons); automatable jobs, where AI can carry out many tasks (e.g. customer service workers, software coders); and less affected jobs (e.g. in agriculture, construction).

Impact studies tend to focus on the potential job losses rather than on the new jobs being created thanks to AI. Studies estimating AI exposure based on potentially automatable tasks tend to over-estimate job displacement. Another approach, which seeks to quantify the observed exposure to AI, is based on the tasks that are actually used in work-related contexts. AI-exposure serves as a proxy to estimate future AI uptake, which is harder to predict as it depends on a combination of aspects including regulation, costs and organisational readiness.

While previous waves of automation mainly affected low-skilled workers and men, exposure to Gen-AI is greater for high-skilled workers and women; this is due to their strong presence in clerical and administrative roles that have a higher risk of automation. Estimates of jobs exposed to AI range from between 5 % and 60 %. The International Labour Organization (ILO) global index of occupational exposure to GenAI estimates that, one in four workers worldwide are in an occupation with some GenAI exposure, and 3.3 % of global employment is in the high-exposure category. The ILO finds significant differences between female (4.7 %) and male employment (2.4 %), which increase with countries' income, as does overall exposure. According to the World Economic Forum's 2025 future of jobs report, demand for the ability to use and manage AI tools (AI fluency) has skyrocketed.

Evolving expectations

With AI evolving continuously, its complexity and limitations are difficult to capture. Recent headlines – for instance, that in large United States (US) tech companies, most tasks performed by white-collar workers will be fully AI-automated within 12 to 18 months; that a bank group in Belgium plans to replace around 1 000 staff with AI within three years; and that AI is a threat to 5 million workers in France – have sparked fears of an AI tsunami sweeping labour markets. While tech firms that build, use and sell AI models also have an interest in highlighting the technology's positive impact, AI tools or pilots are being introduced by all types of companies that expect return on investment.

In its above-mentioned report, the World Economic Forum finds that in 2025, 86 % of employers surveyed expected AI to have the biggest impact on job transformation; 40 % expected workforce reductions due to AI task automation; and 39 % anticipated workers' core skills to change by 2030 (down from 44 % in 2023). The report also predicts that AI-related trends could generate 11 million jobs, while simultaneously displacing 9 million others.¹

Estimates for Europe suggest that 58 % of current work hours could theoretically be automated using existing technologies (agentic AI for cognitive tasks and robots for physical work). Nevertheless, this reflects technological feasibility and is neither a prediction of what is likely to be adopted in practice nor a job-loss forecast.

While AI is expected to increase productivity, the related long-term productivity gains remain uncertain. A 2024 paper based on estimates of exposure to AI and productivity improvements at the task level expected modest macroeconomic effects (a 0.66 % increase in productivity over 10 years), cautioning that even this estimate may be exaggerated, as it captures evidence from easy-to-learn tasks, while effects from hard-to-learn tasks will depend on multiple factors. Other research from 2026 expects a rise in productivity and economic growth in the range of between 1‑2 % and 2‑2.5 %.

While many organisations have adopted some AI applications, most were still in the experimentation or piloting phase in 2025. Investment was accelerating, yet real impact on performance remained elusive. Most executives who considered AI as key to their company's future competitiveness still largely focused on pilots seeking incremental productivity gains, not a deep transformation. Observers have pointed out that firms are cutting jobs because of AI's potential, not its performance, and that despite massive investment, 95 % of enterprise GenAI pilots failed to deliver the expected returns. Behind this disconnect, there seems to be a misunderstanding of how AI creates and redistributes economic value. Experts have suggested that the real value from AI will not so much lie in boosting productivity but in reshaping offers, business models, and market structures – faster than competitors do.

More recently, even among those predicting scenarios with strong AI-induced white-collar job cuts, the narrative is softening towards the idea that AI will reshape more jobs than it replaces. Most reports agree that GenAI can make workers much more productive and create value for companies, while the relative importance of tasks within each occupation is shifting. To reap the benefits of AI, organisations need to prepare the people who work for them, and redesign workflows, rather than merely replace individual automatable tasks. This is consistent with ILO findings that AI's most likely impact will be job transformation.

Early evidence

People tend to adopt GenAI rapidly for private uses, but AI adoption for work purposes is slower. Empirical evidence of GenAI effects is still in its early days, and its impact on total (aggregate) employment has been low for now. While an aggregate effect of AI on employment remains uncertain, sectoral and localised job losses are multiplying.

Existing studies suggest that AI impact on the labour market is mixed and context-dependent. Much hinges on the regulatory framework under which organisations adopt AI tools, and on the work environment into which AI systems are introduced, particularly the presence of other technologies such as algorithmic management and robots. A potentially negative impact of AI introduction can be partially mitigated by labour market institutions and social dialogue accompanying the reshaping of work environments.

Most experts agree that in high-wage and high-skilled occupations, the likely effect would be job augmentation, while routine-based occupations are more prone to job displacement or precarity. Augmentation can result in increased workers' productivity and output quality, for instance in translation, professional writing, financial and legal tasks, consultancy, customer services and software development. However, as mentioned earlier, in case of over-reliance on AI, output quality can also degrade.

GenAI disproportionately affects entry-level jobs, as it can handle tasks such as research, data analysis, report writing and document review. Such jobs can be performed by a senior worker with an AI tool, instead of a junior employee. However, precisely these jobs used to provide training for newcomers and development opportunities. In the US, early-career workers (aged 22 to 25) experienced a 16 % relative employment decline between late 2022 and September 2025, while employment for experienced workers remained stable.² Nevertheless, other research cautions that GenAI exposure is strongly correlated with working from home, and that entry-level hires were falling steeply already before GenAI arrived.

A 2026 analysis of hiring data in the European Union (EU) affirms that occupations with higher AI exposure do not show steeper hiring declines than other roles, and that AI-augmented jobs are holding best. The data show that, since 2023, over 256 000 AI-related jobs have been created in the EU, mostly AI engineers, heads of AI, and AI infrastructure roles.

AI affects working conditions, and can have both positive and negative effects on job quality. Examples on the positive side include time savings, more interesting tasks, stronger work engagement, increased complexity and responsibility, better physical safety, and the potential for job upgrading. Among the negative impacts often mentioned are higher work intensity, reduced autonomy, cognitive under-load (or low mental workload), higher control and monitoring, skills under-utilisation, psychosocial effects, and potentially also job downgrading. While AI increases job demands and resources, algorithmic management changes the way work is done, monitored and evaluated. Workers tend to perceive using AI as a support tool more positively than adapting to AI controlling work processes and monitoring performance.

AI is redefining the skills workers need. A McKinsey Global Institute 2025 report estimates that over 70 % of today's skills can be applied in both automatable and non-automatable work, and predicts that most skills will likely remain relevant but used differently. With the more common tasks handled by AI, workers could spend more time framing questions and interpreting results. Digital and information-processing skills are likely to change substantially, while skills related to assisting and caring could change less.³

The International Monetary Fund (IMF) observes that, while women and college-educated workers are more exposed to AI, they can also reap its benefits more effectively, while older workers are potentially less able to adapt. The IMF warns that AI has the potential to affect income inequality (if the complementarity between AI and high income workers is strong), and that capital returns will increase wealth inequality.

Uneven adoption of AI risks widening economic disparities, both within labour markets and between regions. The Organisation for Economic Co-operation and Development (OECD) observes that GenAI is shifting regional labour market exposure, with regions concentrating industries such as ICT or finance becoming most exposed. In the past, the impact of automation was stronger in semi-urban and rural areas, yet GenAI is impacting higher-skilled jobs in urban areas.

EU perspective

Data collection on AI use is recent and needs to be interpreted with caution. At the individual level, a 2026 consumer survey across 18 EU Member States found that around 54 % of individuals use AI tools. Of these, less than half use AI for work-related reasons, and 91 % of them report time savings thanks to AI use. 'Managers and professionals' report the highest time gains, followed by workers in elementary occupations, students , and people engaged in unpaid work. AI use also yields better work output and reduces work burden. Countries with higher AI adoption rates report lower perceived benefits from AI, and regard the additional benefits as incremental rather than transformative. Among individuals using AI at work, 14 % are 'very' and 27 % 'somewhat' concerned about job displacement due to AI.

At the enterprise level, 2025 Eurostat data show that almost 20 % of all EU enterprises use AI technologies, while use was higher (55 %) among the large ones (compared with 30 % among medium-sized enterprises and 17 % among small enterprises).⁴ This can be explained by the complexity of implementing AI tools, economies of scale, and costs. AI use is more frequent in the information and communications sector and in professional, scientific and technical service activities. In all other sectors, the share of companies using AI was below 25 %. AI uptake has been highest in Denmark, Finland and Sweden, and lowest in Romania, Poland and Bulgaria.

While there was no 'winning AI technology' used by all companies, the large ones mostly use AI tools analysing written language (text mining), followed by AI generating written or spoken language and programming code. The purposes of AI use differ by company size and sector of activity. The main reasons for not considering AI introduction appear to be lack of expertise, lack of clarity about the legal consequences, and data protection and privacy concerns.

A 2025 report for the European Commission notes that, if AI is adopted at pace, up to 6.5 % of the EU workforce may need to transition to new occupations by 2030, as companies will face growing skills mismatches. More generally, it recognises that developing, attracting and retaining AI talent, as well as increasing AI skills more widely, is a defining challenge for EU's long-term competitiveness.

The European Commission put forward its AI continent action plan and Apply AI strategy in April 2025.⁵ The action plan seeks to foster AI adoption in strategic sectors, strengthen AI skills and talent, and facilitate implementation of the AI Act. As regards skills, the plan builds on the 2025 union of skills initiative, announces a future AI skills academy, and intends both to enlarge the EU's pool of AI specialists and to upskill and reskill the EU workforce and population. The Commission also supports continuous learning through the network of European Digital Innovation Hubs, which have been refocused as experience and acceleration centres for AI ('Centres for AI').

The European Parliament, building on a 2022 resolution on the digital divide and the social differences generated by digitalisation, emphasised in a March 2026 resolution that investment in high-quality, accessible education and lifelong learning is a strategic social investment preparing the workforce for the green and digital transitions, including digital and AI literacy. To ensure workers' protection at the workplace, Parliament has called on the Commission to propose a directive that would extend the new rights obtained by platform workers to all workers, and regulate algorithmic management at work.

The 2026 Council recommendation on human capital estimates that the EU will need between 6.2 and 7 million  AI‑related workers by 2027, with around 60 % of the workforce requiring AI skills. It recommends that Member States strengthen digital skills, AI literacy and critical thinking of students, apprentices and lifelong learners, and encourage the effective, responsible, inclusive and ethical use of AI.

A change of focus towards ensuring a skilled workforce is reflected in the 2026 European Semester spring package, which links economic performance with the human capital needed to achieve it. For the first time, country-specific recommendations also focus on quality employment, better functioning of labour markets and skills development, including the development of digital skills, to increase productivity and strengthen social market economy.

Outlook

In these early days of AI adoption, the spectrum of possible developments is broad.

A recent example of a pragmatic approach advises preparing for an AI jobs apocalypse, and suggests preventive measures that governments could adopt to avoid a potential unrest provoked by larger AI-related lay-offs of white-collar workers. These could include slowing down the change (companies adopt AI, but do not lay off workers, as in China); smart tax reforms (levies on excessive corporate profits); and helping workers to adjust and train for new occupations.

AI could bring about a significant income redistribution, away from labour and towards capital. As the occupations most exposed to AI are central to income formation, tax revenues and value creation in advanced economies, the consequences may extend well beyond employment alone. For instance, a rethink of education systems may be needed, as the labour market value of qualifications and diplomas could soon be 'tested through the AI lens', giving more weight to judgement, adaptability and AI complementary skills, for instance technical literacy and analytical skills needed to interpret and supervise AI functioning.

According to more moderate views, AI adoption will take time, and that work will increasingly involve the collaboration between humans and intelligent machines. As AI adoption progresses, some jobs will shrink, others shift or grow, while new ones will emerge. Work may change profoundly, and human skills will be applied differently.

For younger workers between 15 and 24 years of age, employment perspectives are weak, with limited upsides, suggesting that youth employment is most vulnerable to technological developments. Targeted solutions will have to be found and tested. For instance, rather than eliminating entry-level jobs, companies could use AI to train the next generation of senior professionals. Emerging strategies addressing the issue include redesigning entry-level work, reinventing professional development pathways, and collaborating with education and training systems.

Technological change appears to have a skill-biased impact: displacing low-skilled workers, giving mixed outcomes for middle-skilled workers (depending on how tasks are automated), and strongly favouring high-skilled workers. In addition, while high-skilled workers tend to be exposed to AI itself, low-skilled workers are also more exposed to a range of AI‑enabled applications. This combined exposure could magnify the impact on low-skill employment.

For this technological transformation to succeed, the OECD insists on prioritising training and education to provide both current and future workers with the necessary skills. Further key challenges raised by the OECD include adequate social protection for displaced workers, social dialogue, safeguarding workers' rights in the face of AI, and ensuring inclusive labour markets.

Anchoring the debate on AI in its underlying fundamentals, Pope Leo XIV, in his recent encyclical letter, questions the global AI race, warning that the use of AI is not morally neutral. Cautioning that AI could exacerbate inequality, deepen social fragmentation, and weaken moral responsibility if not constrained by ethical limits and democratic oversight, he reflects on what AI should be designed to do. In his view, ethical discernment cannot be limited to asking whether we are using AI for good or for bad purposes – we must also examine how AI is designed, and what vision of the human person and society is embedded in the data and models that guide it.

A longer-term forecast about how AI and digital technologies could reshape employment in European regions is cautiously positive, regarding the adoption and diffusion of AI as a gradual with long-term payoffs. The impact is expected to accelerate over time and become substantial by 2040. Depending on the scenario, the outcomes modelled range from marginal gains, to increases of about 14 % of total employment by 2040. These gains, however, depend on labour mobility and the adaptive capacity of workers, firms and regions. Leadership choices will shape AI adoption outcomes. Policy will play a key role in launching large-scale reskilling and upskilling efforts, youth-focused measures, and national and regional adaptation strategies to translate the technological potential into labour market gains.

In the short-term, the European Commission, building on a December 2025 quality jobs roadmap is working on a quality jobs act. Preparatory consultations suggest that the Commission could seek a balance between supporting and enabling the uptake of AI at work, and protecting workers from potential risks arising from the use of algorithmic management at the workplace.