In establishments that adopt artificial intelligence, the demand for cognitive, emotional, and digital skills is beginning to decline after having increased. This reversal of trend, documented by the OECD on panels of employers, calls into question several years of consensus on reskilling as a response to automation.

The Organization for Economic Cooperation and Development is observing a break in the evolution of skill needs. While a growing number of employers promise internal reskilling of their employees, a significant portion is already preparing workforce reductions. If this dynamic is confirmed, the entire argument for training as a social shield begins to falter.

The Essentials

  • The demand for cognitive, emotional, and digital skills is declining in companies using AI after an initial increase
  • A majority of employers are betting on internal reskilling, but a significant minority is considering job cuts
  • The OECD documents this break in panels of establishments adopting AI
  • This reversal contradicts several years of consensus on training as a response to automation

A Reversal of Trend Documented in the Field

The OECD has tracked cohorts of establishments during their adoption of artificial intelligence tools. The data first shows an increase in demand for cognitive, emotional, and digital skills in job postings. This initial increase corresponded to economists’ predictions: AI would create new complementary skill needs.

Then the curve reverses. In a second phase of adoption, these same skills see their demand decline. The Paris-based organization observes this trend across several sectors without providing a precise timeline, but data collection spans a recent period.

This empirical documentation contrasts with theoretical models that predicted lasting complementarity between AI and human skills. On the ground, automation appears to be moving beyond the augmentation stage and entering that of direct substitution.

The establishments studied primarily cover developed OECD member economies. The sample includes service companies, manufacturing industry, and the public sector, with an overrepresentation of organizations with the means to invest massively in generative AI.

Reskilling Promised by a Majority of Employers

Faced with this evolution, employers are displaying optimism that contrasts with the data on skill evolution. A large majority of executives surveyed by the OECD are betting on internal reskilling of their teams. This proportion varies by sector but remains majority across all studied industries.

Requalification is becoming the central argument of human resources management. Training programs are multiplying, often centered on learning how to use AI tools rather than developing skills that are genuinely complementary to automation.

Singapore has made the individual training account a shield against AI, but this public approach remains the exception. In most OECD countries, training still largely depends on employer initiative.

Yet a significant minority of the companies studied are already preparing workforce reductions. This proportion reveals that not all executives believe their own discourse on reskilling. For these employers, AI represents an opportunity to reduce wage costs rather than a tool for job transformation.

Several Years of Economic Consensus Called into Question

For several years, economists, international organizations, and governments have repeated the same mantra: automation destroys jobs but creates others, provided workers are trained in new skills. This consensus rests on the observation of past technological revolutions, from the steam engine to computing.

Generative AI disrupts this logic. Unlike previous technologies that automated manual or repetitive tasks, it directly attacks the cognitive, creative, and relational skills that economists considered specifically human.

OECD data suggests that this cognitive automation is progressing faster than the creation of new skill needs. The initial phase of augmentation, where AI actually created new jobs and new demands, appears to be closing rapidly.

This evolution calls into question public policies on vocational training across all developed countries. Substantial investments in adult reskilling are based on the assumption that human skills will remain complementary to automation. If this assumption is eroding, the entire architecture of employment policies must be rethought.

The Most Exposed Sectors Accelerate Substitution

The OECD study does not detail sector by sector, but those first affected are predictable. Business services, accounting, consulting, and data analysis are seeing their demand for skilled labor decrease more rapidly than in industry or proximity services.

AI is making inroads into law firms and weakening expertise, illustrating this rapid penetration into intellectual professions. Document research, standard contract drafting, and case law analysis are automating at a speed that surprises even professionals in the sector.

Technology companies, paradoxically, are among the first to reduce their workforces. They master AI tools better and can therefore more precisely measure which positions become redundant. Layoffs in large technology companies since 2023 may prefigure a broader trend.

Conversely, professions requiring physical presence or complex social interactions are holding up better. Education, proximity healthcare, and craftsmanship are maintaining their labor needs, even if AI is transforming some of their tasks.

The Gap Widens Between Promises and Reality

The discrepancy between the intentions stated by employers and the actual evolution of skill needs raises a major political question. If a majority of executives promise reskilling but skill demand is declining, who will finance the training of millions of workers whose jobs are disappearing?

Governments are still largely betting on public-private partnerships and employer commitment. This strategy works as long as employers have an incentive to train their employees. But if AI allows for massive workforce reduction, the economic incentive to train disappears.

France devotes considerable sums to vocational training, much of it financed by companies through the training contribution. This model assumes that employers invest in the skills they need. When these needs collapse, financing reskilling becomes problematic.

Other countries are experimenting with more directed approaches. Germany is studying a “legally enforceable right to training” that would require companies to finance the requalification of employees laid off due to automation. But even these measures assume that substitute jobs exist.

Contradictory Signals About the Future of Work

The OECD is publishing its observations at a time when other indicators are giving contradictory signals. Unemployment remains low in most developed countries, job creation continues, and new professions related to AI are indeed emerging.

This apparent contradiction can perhaps be explained by timing. The massive adoption of generative AI dates only to 2023 with ChatGPT. Its effects on employment are unfolding progressively and first affect the most computerized sectors. Generalization across the entire economy will take years.

The OECD data concerns companies that have already taken the step toward automation. They provide a preview of what awaits less technologically advanced sectors. If this reading is correct, the shock to skilled employment has not yet begun on a large scale.

One major unknown remains: the economy’s capacity for innovation to create new jobs. Economic history shows that technological revolutions destroy professions but invent others. Will AI follow this rule or mark a definitive break? The answer will determine the effectiveness of reskilling policies on which most governments are still betting.

The next OECD reports will be scrutinized by all decision makers. If the trend toward declining skill needs is confirmed and spreads, other answers beyond training will need to be invented to accompany the transformation of work.