Happy Labor Day? How geopolitics, immigration and AI will reshape work

• US and European labor markets appear to be in good shape, with headline unemployment rates near historic lows. But three strong undercurrents (immigration policy, energy-price shock and AI) are churning beneath the calm surface. Demand for workers had already begun to soften before the shock of the Middle East crisis: vacancy rates were easing and hiring rates cooling from uncertainty and the trade war. The shift in immigration policy in the US, the UK and Germany is adding to the mix: Fading immigration inflows have quickly translated into lower employment numbers. In the US,  immigration has gone from contributing over half of job creation in 2024 to near-absent in 2025 – weighing on potential growth and risking sectoral labor shortages, while offering a short-term offset against rising unemployment.
• The Iran shock will land unevenly on labor markets depending on countries’ energy exposure and the duration of the conflict. Still, we see only a limited increase in unemployment rates in the range of 0.1-0.3pp. If the crisis is resolved by end-May, unemployment rates would rise only in the most exposed European economies, and by +0.1pp at most, with 102,000 jobs lost in the Eurozone and half that in the US. Unlike in 2022, when labor hoarding led to falling unemployment, leaner buffers today could see any shock frontloaded into unemployment. In case of a prolonged closure of the Strait of Hormuz, the energy shock would hit Europe harder than the US, though higher labor protection standards provide some cushion: an estimated 225,000 jobs would be lost (0.13% of employment) in the Eurozone versus 126,000 (0.08%) in the US, with higher losses in Germany, Poland, Italy, France and Spain due to high energy exposure.
• AI’s labor-market effects are emerging, pointing to a K-shaped pattern, with youth and mid-level white-collar workers most at risk. Early evidence shows pressure on younger and less experienced white-collar workers in routine cognitive tasks, while gains accrue to higher-skilled, AI-complementary roles. Since late 2022, higher AI adoption has been associated with larger increases in youth unemployment, with AI exposure explaining about 40% of cross-country variation (excluding high-unemployment economies). AI may therefore appear first not as job loss but as fewer entry points, weaker wage growth, and sharper polarization, with reallocation driven mainly by shifts in job composition rather than wage pressures in labor-intensive activities predicted by Baumol.
• In the medium term, the AI labor-market impact will be substantial, uneven across countries and unprecedented in the scale of workforce reorganization required. Over the next 1-3 years, AI is expected to affect 23.3% of jobs across major economies, with reorganization (10.4% of jobs) dominating augmentation (5.3%) and outright displacement (7.6%). The share of jobs affected ranges from 9.2% in Italy to 28.7% in the US, with the UK (17.7%), Germany (16.2%), France (14.7%) and Spain (12.4%) in between. That is equivalent to 52.5mn jobs in the US and 21.8mn across the major European economies.  Our analysis does not account for potential AI-related job growth, which we expect to at least partially offset adverse employment effects. However, job displacement is likely to outpace job creation in the medium term, as firms adjust faster than workers, creating a temporary gap. Ultimately, whether – and how quickly – AI leads to job losses, reorganization, or new job creation will depend less on technology than on policy choices. AI-proofing policy frameworks will be critical: labor-market policies, including re- and upskilling, active labor-market programs, and social protection, will shape worker transitions. Taxation (including the relative treatment of labor and AI capital), firm incentives, and competition policy will determine whether AI is deployed to augment or replace labor and how broadly productivity gains are shared.

Labor markets on both sides of the Atlantic have so far remained resilient despite successive shocks. In the Eurozone, the unemployment rate remained close to historical lows at 6.2% in March 2026, reflecting the job‑rich post‑Covid recovery and the strong employment response relative to modest growth. In the US, labor‑market conditions have been weaker, with net job creation grinding to close to zero since early 2025, after dynamic outturns in 2024. However, stalling job creation mostly reflects sharply weakening labor supply amid very tight immigration policy, though plunging labor demand for federal employees has contributed, too. As a result of lower immigration inflows, the US unemployment rate has not crept up by much: from 4.1% in January 2025 to only 4.3% in March 2026.

AI is altering how firms adjust to the energy shock, not replacing energy costs as its source. With AI already embedded in production processes, firms have gained greater scope for internal substitution, particularly in routine cognitive tasks and entry‑level and mid‑level white‑collar roles . In a context of elevated energy costs and softer demand, this lowers the threshold at which firms move from margin and working‑time adjustment to labor reallocation, primarily through hiring restraint and weaker entry‑level absorption rather than outright job destruction. Energy prices determine the need to adjust; AI shapes the speed, intensity and distribution of that adjustment.

Accordingly, more AI‑intensive sectors have displayed muted or zero employment growth in the Eurozone since late 2022, consistent with adjustment taking place through subdued hiring and task reorganization and slower entry-level inflow rather than outright job destruction. This reinforces the view that AI dampens measured employment dynamics even as underlying restructuring progresses.

From a Baumol perspective, uneven productivity growth tends to generate upward pressure on relative wages in less automatable, labor‑intensive activities. In the current environment, however, this channel is muted: with AI dampening aggregate labor‑demand growth and firms adjusting primarily through hiring restraint and task reorganization, rebalancing is taking place mainly through employment composition rather than prices. Wage adjustment is therefore asymmetric, with compression in AI‑exposed white‑collar and entry‑level roles, while less‑AI‑intensive sectors absorb labor without broad‑based wage acceleration.

AI has not yet raised aggregate unemployment, but it is already reshaping labor‑market adjustment in a k-shaped manner. Since late 2022, countries with higher AI diffusion have experienced larger increases in youth unemployment, with AI exposure explaining around 40% of the cross‑country variation (excluding economies with structurally high youth unemployment), even as overall unemployment rates remain contained. At the same time, countries with comparable levels of AI uptake display markedly different labor‑market outcomes, underscoring the decisive role of institutions, labor‑market structures and adjustment channels. This divergence suggests that AI diffusion does not mechanically translate into net job losses, but neither is it benign: it shapes who adjusts, how quickly and at which margin, depending on how easily firms can substitute tasks, how flexibly workers can be reallocated and how effective retraining and activation policies are. Where adjustment mechanisms are rigid and transition support is weak, the risk is not an immediate rise in aggregate unemployment but persistent scarring at the margin, concentrated among younger workers and in routine‑intensive entry‑level occupations. The policy challenge therefore lies less in protecting aggregate employment than in shortening transitions and reducing polarization, through portable skills, flexible wage‑setting and active labor‑market policies that allow economies to capture the productivity upside of AI without incurring lasting distributional costs.

As AI moves from initial deployment to broader integration in business processes, the focus shifts to how it shapes labor markets in the near- to medium-term. While it remains difficult to predict how the technology will evolve and diffuse, the next one to three years are likely to be critical in determining the extent to which current capabilities translate into tangible changes in jobs and tasks. Against this backdrop, we assess the potential impact of AI on the current workforce across the Eurozone’s four largest economies – Germany, France, Italy and Spain – as well as the UK and the US.

AI job exposure varies significantly across sectors. As a starting point for the analysis, we identify AI exposure across sectors, measuring the share of tasks that AI can perform significantly faster or more efficiently. Based on the recent analysis of Falck and Röhe (2026) it ranges from around 14% in agriculture and 22–25% in construction and manufacturing to over 40% in finance and real estate and nearly 48% in information and communication. These differences in AI exposure are driven by task composition and are broadly stable across countries. Sectors dominated by cognitive work are substantially more exposed, while manual and physical tasks remain less affected.

Cross-country labor market exposure to AI is driven by economies’ sectoral employment composition. While AI affects sectors similarly across countries, economies differ in how their workforce is distributed across those sectors. Translating sectoral exposure into country-level outcomes therefore depends on national employment structures, making economic composition the key driver of aggregate exposure. For example, high-exposure sectors such as ICT (48%), finance and real estate (41%), public administration and defense (26%), and health and education (15%) account for a larger share of employment in the US and partly the UK than in continental Europe. By contrast, Italy and Spain are more concentrated in lower-exposure sectors such as manufacturing (18% and 12%) and agriculture (3%), reducing aggregate exposure. Germany occupies a middle position among all countries, combining a sizable manufacturing base (18%) with more moderate service exposure. As a result, even with identical sectoral exposure, total AI impact differs substantially across countries, with service-oriented economies more exposed than industrial or agriculture-based ones.

Among AI-exposed jobs, we apply a job-transition framework that distinguishes between three adjustment channels: augmentation, reorganization and automation. These categories capture how firms are likely to restructure tasks within the existing workforce rather than simply whether jobs disappear. Based on the analysis by Martin Richmond (2026), 22% of AI-exposed jobs are expected to be augmented, 44% reorganized and 33% automated. Augmented jobs expand as productivity gains stimulate labor demand, reorganized jobs retain their core function but undergo task-level changes while for automated jobs humans are fully replaced by AI.

How these job-level effects translate into labor-market outcomes depends on country-specific conditions, which determine how quickly and to what extent AI exposure materializes over the next 1–3 years. We capture this through four scaling factors that amplify or cushion the impact. In the short term, AI adoption and workforce readiness for AI determine how quickly theoretical exposure causes actual impact. Countries such as the US and the UK lead both in current AI adoption and workforce readiness for AI, while Italy ranks the lowest among the countries considered, with Germany, France and Spain placed in the middle. In addition, we consider more structural factors such as unit labor costs and employment protection legislation that will shape firm incentives in adopting AI as well as influencing whether AI is used to complement or replace workers. High labor costs clearly increase the incentive to automate, while stronger employment protection legislation tends to slow displacement, at least in the near- to medium-term. For instance, the US has the weakest labor market regulations among the countries considered, leading to a higher expected pass-through of the theoretical AI exposure into actual labor market impact. By contrast, Italy has the strongest employment protection legislation, which shields the economy from the AI-related labor market disruption at least temporarily but also limits the economy’s ability to benefit from AI-enabled productivity gains.

The AI impact on the current workforce across major European economies and the US is substantial but far from uniform. To gauge the overall AI impact on the current workforce, we combine multiplicatively at the country level the sectoral AI exposure estimates, sectoral employment structures as well as the four scaling factors. The results are country-specific shares of jobs affected by AI, reflecting both technological potential and real-world constraints. Over the next 1–3 years, AI is expected to affect 23.3% of jobs across major economies, with reorganization (10.4% of jobs) dominating augmentation (5.3%) and outright displacement (7.6%). The share of jobs affected ranges from 9.2% in Italy to 28.7% in the US, with Spain (12.4%), France (14.7%), Germany (16.2%) and the UK (17.7%) in between. These differences closely mirror the distribution of employment across sectors. Economies with larger shares of finance, ICT and professional services exhibit substantially higher exposure than those with more manufacturing, construction or agriculture.

Across all countries, the most important adjustment channel is job reorganization, while automation remains more material than augmentation. This reflects the fact that in most AI-exposed jobs, a human in the loop remains necessary, leading firms to reorganize tasks rather than replace workers, while only a smaller share of jobs meets the conditions for full automation and even fewer benefit from demand-driven expansion. In the US, 12.9% of jobs fall into the reorganization category, compared to 7–8% in Germany and the UK and 4–6% in Southern Europe. This finding is consistent with AI adoption currently targeting auxiliary or “side” tasks rather than core functions, reinforcing a complementary rather than substitutive dynamic. The share of jobs at risk ranges from 3.4% in Italy to 9.0% in the US, with most countries clustering around 5–6%. This gap between exposure and actual automation reflects the role of labor-market regulation, human-interaction requirements and demand effects, which limit the immediate substitutability of labor even where tasks are technically automatable. With respect to the expected real world labor impact of AI over the next 1-3 years, between 2% and 7% of jobs benefit from AI-driven expansion, thanks to augmentation, with the strongest effects in the US and UK. These gains arise where productivity improvements outweigh labor costs and hence increase employment.

The transatlantic divide in AI labor-market exposure is substantial and systematic: the US can expect more short-term AI-related labor market pain coupled with more long-term gain. In the US, 28.7% of jobs – equivalent to around 52.5mn positions – are affected by AI, compared to 14.5% (21.8mn jobs,) across the major European economies (Germany, France, Spain, Italy, UK). This pattern largely reflects structural differences in economic composition. The US economy is more heavily weighted toward high-exposure sectors such as information, finance and professional services, where AI can be deployed at scale, whereas Europe – particularly Southern Europe – retains a larger share of lower-exposure sectors such as manufacturing, construction and hospitality. The comparison between the US and Italy illustrates the extremes of this distribution: while the US combines high exposure (28.7%) with strong augmentation (6.9%), reorganization (12.9%) and automation (9.0%), Italy shows low exposure (9.2%) and correspondingly limited adjustment. As a result, the US faces greater transition pressures but also stronger potential productivity gains, whereas Europe’s lower exposure provides short-term stability but risks slower productivity growth and weaker AI-driven job creation over time.

A key limitation of these results is that they focus on the current workforce and do not account for AI-induced job creation. Our estimates should therefore not be interpreted as direct changes in headline employment. Historically, technological change has followed a consistent pattern: job destruction by automating tasks is offset over time by job creation through new tasks and demand effects. However, this adjustment is neither immediate nor evenly distributed. Job creation typically lags behind displacement due to reallocation frictions and skill mismatches. The time mismatch may be more pronounced for AI, as diffusion and adoption rates are unprecedented. Taken together, AI-driven job losses are likely to be partially offset over time, but with meaningful transition costs.

The scale of required job creation is material. Our results indicate that around 25mn jobs across the US and major European economies are subject to automation. Fully offsetting displacement would therefore require the creation of a comparable number of new jobs over the medium term. Early evidence shows strong growth in AI-related roles with some 1.6m new AI-related jobs like AI engineers, forward-deployed engineers and data annotators created over the past two years plus 600,000 jobs driven by the data center boom with the US the key beneficiary. However, diffusion across sectors and countries remains uncertain. Net employment outcomes will depend on the speed of job creation relative to displacement and on the efficiency of worker reallocation. In ageing economies, declining labor supply may additionally cushion the employment impact as labor shortages persist in many sectors.

AI does not determine labor-market outcomes on its own. The technology sets the direction, but policy and institutions determine the speed, scale and distribution of its effects including the balance between job augmentation, reorganization and automation. What distinguishes the current wave of AI is not only its breadth across sectors, but the pace and depth of task-level transformation, amounting to one of the most significant labor-market transitions in recent decades. Existing policy frameworks, designed for slower, occupation-based change, are therefore ill-suited to the AI age, where jobs are continuously reconfigured. To remain effective, education systems, active labor-market policies, social protection systems, tax structures and firm-level incentives must be updated and rethought to become more forward-looking. The main challenge is therefore not preventing adjustment, but to shape its direction and distribution, ensuring that productivity gains translate into broad-based improvements in employment, wages and job quality rather than displacement or increased inequality.

First, labor-market policy will determine how smoothly workers move between tasks, roles and sectors. Traditional active labor-market policies will need to evolve from reactive systems focused on unemployment toward anticipatory, data-driven and task-based approaches. This includes early-warning systems to identify AI exposure, individual learning accounts, and rapid retraining mechanisms that can be deployed before displacement occurs. Public employment services should be strengthened with real-time labor-market intelligence and AI-enabled matching tools that connect workers to adjacent roles based on transferable skills. Wage insurance, portable benefits and unemployment support linked to retraining will be essential to reduce income risk and support mobility. Since most AI-exposed jobs are likely to be redesigned rather than eliminated, the priority is not protecting existing tasks, but enabling workers to transition quickly into new ones. Since most AI-exposed jobs are likely to be redesigned rather than eliminated, the priority is not protecting existing tasks, but enabling workers to transition quickly into new ones.

Second, education and skills will become more binding constraints. As AI increasingly complements cognitive work, the ability to work alongside AI systems will determine whether workers benefit from productivity gains or face substitution. This requires embedding AI literacy as a core competence across education systems, alongside stronger emphasis on critical thinking, problem-solving, adaptability and communication. Education systems will need to become more modular and continuous, with expanded mid-career training, short-cycle credentials and closer alignment with labor-market needs. Governments should also support sector-specific training ecosystems and local partnerships between firms, unions and education providers to accelerate adjustment in regions and occupations most exposed to disruption.

Third, firm incentives, including taxation and regulations will shape whether AI is used to replace labor or to augment it. A key issue is that labor is typically taxed more heavily than capital, which creates a structural bias toward automation even when augmentation would be more socially efficient. Tax systems therefore need to be adjusted to incentivize augmentation rather than the substitution of labor. One proposal is a “robot tax,” aimed at reducing this bias by taxing automation more in line with labor—for example, by limiting preferential tax treatment of capital that substitutes for workers or linking levies to displacement. Implementation challenges would need to be addressed, including defining what constitutes a “robot,” avoiding disincentives to innovation, and limiting adverse competitiveness effects if policies are not coordinated internationally. Whereas broad-based taxation of automation may be difficult to implement in practice, more targeted approaches could be more effective. These include tax incentives to support augmentation, such as credits for training and reskilling, incentives for job redesign, and support for firms investing in human capital alongside technology. Frameworks can also encourage sharing of productivity gains through wages, profit-sharing, or reduced working time. Policies to support AI adoption among SMEs, strengthen competition, and prevent excessive concentration of productivity gains will also be critical. Even small differences in these incentives can lead to large differences in business and labor-market outcomes.

Finally, the broader societal response will matter: Public concerns around economic disruption, trust and safety, and the distribution of gains may lead to calls for stricter AI regulation and shape how AI diffuses across economies. Maintaining social legitimacy will therefore be critical. This requires stronger transparency in the use of AI in the workplace, appropriate safeguards around its deployment, and mechanisms for worker voice and participation in how AI is introduced and used. Ensuring that the gains from AI are broadly shared will also be central to sustaining support for adoption. This includes profit-sharing mechanisms and competition frameworks that prevent excessive concentration of AI-driven rents. A complementary priority is to ensure that AI adoption is broad-based rather than concentrated. This calls for policies that lower barriers to adoption, including support for SMEs, promote competition, and facilitate the diffusion of AI capabilities across firms and sectors. Such measures will be critical to ensure that productivity gains translate into wider economic benefits and employment growth.