European Union Continental AI Action Plan: Analysis of the Portuguese Labor Market
European Union Continental AI Action Plan: Analysis of the Portuguese Labor Market
The complete study is available for download here.
Important note: The “word” file must be opened before clicking for the first time in the “excel”.
A página em português para este Projeto pode ser consultada aqui.
The European Commission adopted a Continental Action Plan for Artificial Intelligence (AI), designed to help position the EU as a global leader in this technology. The Plan is built around five pillars intended to accelerate AI adoption, stimulate innovation, and provide greater regulatory clarity within the European framework. In this evaluation, we focus on pillars (3) and (4), respectively:
- AI adoption in strategic sectors: promotion of AI use in the public and private sectors, particularly among SMEs, through an integrated strategy that combines support instruments, skills development, innovation infrastructure, and European funding.
- Development of AI skills and talent: strengthening the supply and attractiveness of AI education and advanced training by expanding academic programs, attracting international talent, and supporting SMEs and the public sector through European Digital Innovation Hubs and European skills development programs.
Objectives
Position the European Union as a global leader in artificial intelligence by promoting broad, competitive, safe, and innovative adoption.
Final assessment of the measure:
Probably effective.
Intergenerationally unfair in the absence of active labor market policies (ALMPs) that correct the distributive effects of AI adoption – not because of the policy itself.
Policy Context and Analytical Approach
Historically, the European Union’s approach to artificial intelligence has been preventive and ethics-oriented. That regulatory emphasis, however, has been criticized for potentially constraining innovation and competitiveness relative to more flexible economies such as the United States and China. The productivity gap between the EU and the United States over the past decade is often associated with the greater ability of the U.S. economy to adopt new technologies quickly (Draghi, 2024). At the same time, the EU faces important constraints in training and retaining AI talent. Although it trains a meaningful share of the field’s leading researchers, only a minority remain working in Europe, reflecting wage differentials and fewer opportunities for funding and firm scale (World Economic Forum, 2025). These weaknesses are also reflected in lower AI use among European workers compared with U.S. workers (Sukharevsky et al., 2024), and in a more limited presence of European universities in top data science and AI rankings.
To address these challenges, the European Commission defined a strategic plan aimed at accelerating the adoption of artificial intelligence and strengthening the skills and talent base needed for European competitiveness.
In this context, this paper proposes an intergenerational evaluation of the effects of AI diffusion on the labor market and of the role of the European Commission’s strategic plan in that process. The analysis focuses on the Portuguese context, identifying the groups most exposed to technological change and the potential implications for inequality between and within cohorts over the working life cycle. Finally, it presents public policy recommendations, with particular emphasis on the role of active labor market policies (ALMPs) in mitigating the most adverse distributive effects.
Artificial Intelligence in the Portuguese Labor Market
The analysis of the Portuguese labor market is based on a task-based approach, which allows occupations to be classified according to their degree of exposure to AI-related substitution and complementarity (see Methodological Appendix). Based on this classification, four types of occupations can be distinguished, each with a different profile of risk and opportunity in the face of AI diffusion. First, there are occupations with low exposure to substitution, generally linked to non-routine manual tasks and in-person services. Second, non-routine cognitive occupations tend to benefit from greater complementarity with AI, with potential productivity gains. Third, occupations based on routine tasks, whether cognitive or manual, are more vulnerable to automation. Finally, there is a group of hybrid occupations that combine substitution risk with complementarity potential, typically associated with technical and specialized roles.
The evolution of the composition of the Portuguese labor market shows a persistent predominance of occupations with low substitution risk and limited potential for complementarity with AI, reflecting the structural weight of manual activities and services that are less exposed to technological change. At present, more than 80% of employment remains concentrated in occupations with low substitution risk, suggesting that direct exposure of the Portuguese labor market to AI-driven transformation is still limited in the short run (see Methodological Appendix). Even so, there is a gradual shift toward occupations with greater potential for technological change, particularly roles with high complementarity with AI, which are associated with higher levels of education and higher wages, pointing to a gradual transition toward higher value-added activities. By contrast, the occupations most vulnerable to substitution remain concentrated in routine tasks and in the more exposed segments of the labor market, showing that the structural reconfiguration induced by AI is unfolding gradually but with uneven effects across occupational groups.
Impacts and Risk Groups
The distribution of employment shows that a non-negligible and growing share of Portuguese workers is in occupations with high substitution risk and low complementarity potential. Even so, the national labor market remains predominantly concentrated in professions with low substitution risk and limited complementarity potential.
There are, however, important differences in the composition of employment. Women are more concentrated in occupations with higher substitution risk, associated with routine cognitive tasks and therefore more susceptible to automation, while men predominate in roles that combine substitution risk with greater potential for technological complementarity. Occupations with lower substitution risk show greater gender balance. This distribution suggests that AI diffusion may have differentiated effects by gender, with the potential to aggravate pre-existing vulnerabilities if adequate mitigating policies are not adopted. Foreign workers, by contrast, are more heavily represented in occupations with low substitution risk and limited complementarity potential, while their distribution across the remaining categories is relatively even.
In terms of qualifications and pay, occupations with greater complementarity potential tend to concentrate workers with higher education levels and higher wages. By contrast, occupations with lower complementarity are associated with lower levels of qualifications and pay.
In summary, the distribution of employment suggests asymmetric exposure to the effects of AI, allowing two particularly vulnerable groups to be identified in the Portuguese context:
- Workers at the beginning of their careers, currently transitioning into the labor market, who enter routine cognitive occupations. This group is particularly vulnerable to technological substitution because AI more easily replicates tasks based on formal and codified knowledge, which depend less on accumulated experience and contextual judgment. The rise in unemployment among recent college graduates in the United States suggests that this phenomenon may already be underway.
- Workers in more advanced stages of their careers with lower levels of education, employed in routine roles with limited potential for technological complementarity. This group is highly vulnerable to substitution, not only because of the nature of the tasks they perform, but also because of the growing automation of cognitive and manual activities driven by the joint diffusion of artificial intelligence and robotics. They also face greater difficulty adapting to new technological demands, owing to lower levels of formal education and less access to lifelong learning opportunities.
Interactive Occupation Map
Each rectangle represents an occupation in 2022. The area of each rectangle is proportional to the number of workers, and the color reflects the substitution index, from green (lower) to red (higher). You can sort by different metrics and filter by gender, region, education, sector, and quadrant.
Assessment of the Policy’s Intergenerational Fairness
This assessment focuses on pillars (3) and (4) and is therefore limited to those two dimensions. As a counterfactual, we consider an alternative scenario in which the policy is accompanied by ALMPs intended to mitigate its potential impact on the labor market.
- Increases inequality between generations? Probably yes.
Workers in more advanced stages of their careers tend to benefit less directly from AI in the short run. However, this also means they are relatively more protected from substitution risk. The problem is concentrated above all among early-career workers trying to enter the labor market: they are the ones facing greater exposure to technological substitution, especially in access to employment. The most likely result is therefore an increase in inequality between cohorts. In the long run, although the exact magnitude of these effects remains uncertain, there are strong reasons to expect them to persist, in line with the scarring effect, that is, the lasting effects of starting working life during a period of structural transformation. This is the expected path in the baseline scenario, that is, in the absence of ALMPs. With adequate complementary instruments, these effects could be significantly mitigated. - Increases intragenerational inequality? Probably yes.
The effects of this measure will not be homogeneous within each cohort. Among early-career workers, the impact will depend above all on the type of occupation, namely the degree of exposure to substitution and the potential for complementarity with AI. Although initial socioeconomic status does not by itself determine that risk, it continues to shape it in relevant ways through education, field of study, and access to opportunities. For this group, the effect will therefore be unavoidably unequal. Among workers in more advanced stages of their careers, the risk of worsening intragenerational inequality is even clearer. Substitution is concentrated in routine cognitive occupations, which tend to be associated with lower wages and fewer opportunities for retraining. Rather than reducing these differences, the policy, in the absence of ALMPs, may reinforce them within the cohort itself. - Reinforces the transmission of inequality across generations? Probably yes.
In the absence of ALMPs, and given that short-run impacts are unevenly distributed across age groups, it is reasonable to expect that these inequalities will not only persist but also be transmitted across generations. Workers at the beginning of their careers and transitioning into the labor market are likely to be the most penalized. On the one hand, they face the scarring effects of entering working life during a period of deep technological change. On the other hand, they were largely trained in a pre-AI context and therefore may not possess the skills that are likely to be more valued in the future. This vulnerability is even greater when they enter routine cognitive occupations, which are more exposed to technological substitution. In a counterfactual scenario with effective ALMPs, these effects could be significantly mitigated. - Limits the choices of future generations? Probably not.
In the short run, there may be some reduction in the choices available to future generations. Even so, in the long run the impact is uncertain, and it is not expected that those choices will diminish, given the evidence associated with previous technological revolutions.
Final assessment: Probably effective. Intergenerationally unfair in the absence of active labor market policies (ALMPs) that correct the distributive effects of AI adoption – not because of the policy itself.
Recommendations: Complement the Plan in Education and Labor
So far, the Plan’s focus has been above all on ensuring that Europe has the talent needed to lead this technological
transformation. That objective is important, but it is clearly insufficient. Widespread adoption of artificial intelligence will have profound effects on the labor market, and those effects can no longer be treated as a secondary issue in the design of public policy. By automating tasks and affecting different occupations and age groups unevenly, this transformation requires complementary instruments that mitigate distributive risks and effectively strengthen workers’ ability to adapt (Brynjolfsson et al., 2025; del Rio-Chanona et al., 2025).
At the same time, in the area of education and training, the diffusion of AI requires a clear reorientation of the skills that are valued. This reorientation is not limited to digital literacy or the functional use of AI tools. It also requires a deliberate strengthening of transversal skills that are hard to substitute, namely critical thinking, communication, and the ability to adapt to dynamic contexts (Felten et al., 2023; Hutson & Ceballos, 2023).
Against this backdrop, we recommend the adoption of complementary policies that explicitly strengthen the labor and training dimensions of the Plan, ensuring a more balanced, more robust, and more sustainable technological transition:
- In education:
• Curricular strengthening in schools and universities, including teaching units that equip young people to use AI tools throughout their academic path and in future professional contexts, making use of the expected gains in efficiency and productivity.• Greater emphasis on the development of transversal skills in primary, secondary, and higher education curricula.• Promotion of vocational education and training aligned with regional labor market needs and directed toward manual and non-routine occupations with high structural demand and lower susceptibility to automation. - In the labor market:
• Strengthening active labor market policies, with broader coverage and explicit targeting of the groups most vulnerable to technological substitution, including mechanisms for early identification of occupations at risk and structured job-to-job transitions in line with regional labor market needs. For example, the Portuguese government could create a program similar to Estágios +Talento, but targeted at recent college graduates seeking occupations with tasks at high risk of AI substitution. However, such a program would have to last longer than six months, since it is implausible that in such a short period a worker would acquire enough experience to move into tasks with lower risk of AI substitution.• Expansion of training programs in digital and AI-related fields, ensuring greater scale and better alignment with labor market needs for both employed and unemployed workers, while complementing specialized training with intermediate qualification pathways.
Methodological Appendix: Classification of Occupations and Exposure to AI
The analysis of the Portuguese labor market is based on a task-centered approach that combines information from
O*NET – an occupational database developed by the U.S. Department of Labor, with detailed information on the tasks and skills associated with each profession – with microdata from Quadros de Pessoal to estimate, in Portugal, the exposure of different occupations to substitution and complementarity by artificial intelligence. Occupations were classified according to the type of tasks they involve and distributed across four analytical quadrants that combine high and low levels of substitution and complementarity:
- LS-LA corresponds to low substitution by AI and low complementarity with AI;
- LS-HA, low substitution and high complementarity;
- HS-LA, high substitution and low complementarity;
- HS-HA, high substitution and high complementarity.
These results were then mapped to the Portuguese CPP 2010 classification and applied to longitudinal worker data (Quadros de Pessoal), making it possible to identify the occupations that occupy the four occupational quadrants in the Portuguese labor market, as illustrated in Figure 1.
For more information on the methodology used, please see the main study.
Figure 1: Composition of the Portuguese labor market by main task type in 2022
(aggregated at the occupation level)

Sources: O*NET, Quadros de Pessoal, and authors’ calculations.
Publishing Date: 2026-03-25
About Instructors
