A landmark study of 923 occupations across nearly 30 countries reveals that AI is reshaping the workforce in ways that aggregate employment statistics are almost entirely failing to capture.

by Kasun Illankoon, Editor in Chief at Tech Revolt

[For more news, click here]

The unemployment rate looks fine. Job postings are still being filled. The economy, by most conventional measures, appears to be absorbing artificial intelligence without visible distress. And yet something is shifting beneath the surface of the labor market in ways that the headline numbers - the figures governments report, the ones that make it into quarterly economic reviews - are almost entirely failing to capture.

That is the central and uncomfortable finding of a major new joint study by Coface, the global credit insurance group, and the Observatory of Threatened and Emerging Jobs (OEM). Released in 2026, the report - covering 923 occupations across nearly 30 countries - offers what may be the most granular mapping yet attempted of how AI-driven automation is actually distributed across the working world. And its conclusions challenge some of the most widely held assumptions about which workers are safe, which are not, and why the official data keeps telling us everything is fine when the reality on the ground is considerably more complicated.

Why the Statistics Are Missing the Story

To understand the gap between what the data shows and what is actually happening, it helps to understand how labor market statistics are constructed.

Aggregate employment figures measure headcounts. They tell you how many people are employed in a given sector or occupation at a given moment. What they do not tell you is how the content of those jobs is changing - which tasks within a role are being quietly automated, which responsibilities are shifting, and which entry points into a profession are disappearing even as the overall employment number for that profession holds steady.

This is precisely where the Coface and OEM methodology makes its most important contribution. Rather than analyzing occupations as monolithic units, the study breaks each of the 923 professions examined down into individual tasks, and those tasks into what the researchers call "elementary actions" - described as triplets of verb, object, and context. Each of those actions is then scored for its exposure to automation using explicit, reproducible rules that can be audited and updated as AI capabilities develop.

The result is a forward-looking framework that projects exposure across five distinct phases of AI development, rather than offering a single snapshot frozen at one moment in time. It is, in methodological terms, a significant advance on analyses that rely on expert judgment or on AI systems evaluating their own displacement potential - both approaches that carry obvious limitations.

What this granularity reveals is that AI's impact on the labor market is not evenly distributed across sectors or even across occupations. It is distributed across tasks. And the tasks now most at risk are not the ones that previous waves of automation targeted.

The Shift That Changes Everything

For decades, the conventional wisdom about automation held that routine, repetitive, physical tasks were the most vulnerable - the assembly line worker, the data entry clerk, the toll booth operator. Cognitive work, particularly complex and non-routine cognitive work, was considered the relatively safe end of the spectrum. A lawyer analyzing a novel contract, an engineer designing a new system, a financial analyst modeling a complex scenario - these were the roles that technology was expected to augment rather than replace.

AI breaks that assumption. The Coface and OEM study is explicit on this point: AI does not represent a continuation of robotics or conventional software automation. It represents a categorical shift in the frontier of what can be automated, moving that frontier directly into the cognitive, complex, and skilled functions that have historically been the most secure.

In the study's primary scenario - focused on the deployment of agent-based AI systems capable of autonomous multi-step reasoning - approximately one in eight occupations crosses a threshold of 30% automatable tasks. The study identifies this as the point at which a profession faces profound transformation: not necessarily disappearance, but significant restructuring of its content and, potentially, its workforce requirements.

The occupations clustering above that threshold are concentrated in fields that are highly cognitive and information-intensive: engineering, information technology, administrative management, finance, law, and certain creative and analytical professions. These are not marginal roles. These are the occupations that sit at the center of modern knowledge economies, that generate disproportionate shares of tax revenue and economic value added, and that have historically served as the primary pathway to middle-class economic security.

The least exposed occupations, by contrast, are predominantly manual or involve human interactions that resist standardization: manufacturing, construction, maintenance, transport, food service, cleaning, and care work. The professions most protected from AI disruption, at least in this phase of development, are often the ones that command the lowest wages and the least social prestige.

The Numbers That Demand Attention

The study's quantitative findings deserve to be read carefully, because they carry significant implications that extend well beyond the labor market itself.

More than a quarter of all work content could be automated in four occupational groups: management and administration, creative professions, law and finance, and engineering and IT. That is not a marginal share. It represents a structural shift in what these professions require of the people who hold them.

At the country level, exposure ranges from approximately 12% of total work content in Turkey to nearly 20% in the United Kingdom. The pattern is consistent with the underlying economic structure: the wealthiest, most service-oriented, and most cognitively intensive economies show the highest exposure. The Netherlands, Ireland, and Luxembourg sit alongside the UK at the upper end of the distribution. Countries where employment remains more concentrated in trade, construction, transport, and physical services show more moderate exposure.

This is a finding with geopolitical texture. The nations that have spent decades building competitive advantage through the development of high-skill knowledge economies may be the same ones most structurally exposed to the current phase of AI deployment. The countries that industrialized later, or that retain larger shares of employment in manual and service sectors, may paradoxically find themselves with lower near-term automation exposure - not because they are technologically behind, but because their workforce composition happens to align less directly with what current AI systems can automate.

The Infrastructure Bet That the GCC Is Making

Against this global backdrop, the strategic posture of the Gulf Cooperation Council countries takes on particular significance. Saudi Arabia and the UAE are not waiting to see how AI reshapes the labor markets of others. They are making sovereign-level investments in the infrastructure that will determine who controls AI capacity in the first place.

Photo: Mohamad Jomaa, CEO and Country Manager for GCC and Egypt at Coface

Mohamad Jomaa, CEO and Country Manager for GCC and Egypt at Coface, framed the regional commitment in terms that go beyond the headline investment figures: "The UAE and Saudi Arabia are moving decisively to position themselves as global AI and compute hubs. Commitments measured in the tens of billions of dollars - including multi-gigawatt data center campuses, advanced GPU capacity, and national AI platforms - signal a clear understanding that AI competitiveness starts with infrastructure. These are not pilot initiatives; they are long-term, sovereign-level investments designed to anchor economic diversification and global relevance in a data-driven world."

The logic embedded in that strategy is worth unpacking. If the Coface and OEM analysis is correct that AI will most significantly disrupt the cognitive, knowledge-intensive occupations at the heart of advanced economies, then the countries best positioned are not necessarily those with the largest existing stocks of those occupations, but those with the infrastructure to develop, deploy, and govern AI systems - and the policy frameworks to manage the transition for their workforces. The GCC's investment posture reflects a calculation that being a producer and host of AI infrastructure is a more durable competitive position than being a consumer of it.

The Questions That Go Unanswered - and Why That Matters

One of the study's more important intellectual honesty moves is its explicit acknowledgment of what it does not measure. The exposure scores it generates are, by design, supply-side assessments. They measure the technical potential for automation, not the actual pace at which automation will occur, the new tasks that may be created, or the friction - regulatory, cultural, organizational - that may slow deployment.

Net job losses are not inevitable, and the study does not claim otherwise. What it does claim is that the transformation of job content is already underway, that it is concentrated in places that aggregate statistics are poorly equipped to detect, and that the downstream effects extend far beyond the labor market itself.

Three broader consequences stand out. First, if AI automates a significant share of tasks in the highest-skilled and best-compensated occupations, it risks shifting a substantial portion of economic value added from labor to capital - with direct implications for tax revenue, social insurance systems, and the fiscal capacity of governments to manage the transition.

Second, the relationship between educational attainment and labor market outcomes may weaken in ways that are difficult to predict but significant to prepare for. If the tasks for which long educational pathways prepare students become more easily automatable, the premium that employers have historically placed on formal qualifications may compress. The attributes that remain complementary to AI - judgment, adaptability, the capacity to supervise and interpret AI outputs - may not be the ones that current educational systems are best designed to cultivate.

Third, the concentration of AI's most critical assets - semiconductors, large language models, data center capacity - among a small number of companies and countries introduces new geopolitical and operational vulnerabilities that have no precedent in previous waves of technological disruption.

The Coface and OEM study is careful not to be fatalistic. The transition from technical exposure of tasks to net effects on employment is neither automatic nor linear. History offers numerous examples of technologies that appeared threatening to existing occupational structures but ultimately generated as many roles as they displaced, even if those roles looked very different from their predecessors.

What history does not offer is a precedent for a technology that simultaneously targets the most cognitively demanding, highest-value, and most economically central occupations across the full range of advanced economies at this pace and at this scale.

The aggregate statistics will catch up eventually. The question is whether the institutions - governments, employers, educators, social protection systems - will have moved fast enough by the time they do.