The dominant narrative about artificial intelligence and the labour market has been very clear from the start — and frightening: machines learn to do humans’ work and jobs disappear. But lately there have been doubts about that, and when reviewing the studies and reports published in 2026, the picture that emerges is very different from that catastrophist headline.

Based on the available data, at least for now, the most reasonable and defensible conclusion is not aggregate net massive job destruction caused by AI — at least not yet — but rather a reconfiguration of the labour market, accelerated, to be sure, in terms of tasks, skills, productivity and changes in demand for labour profiles.

It is worth paying attention to a couple of nuances: yet and aggregate. In other words, the 2026 data still do not detect a net loss of employment across the economy as a whole.

What the employment data show — and what they do not

The most continuous monitoring is carried out by the Yale Budget Lab, which tracks the impact of AI on the US labour market, updating its analysis every month with each new release of the Current Population Survey (CPS). Its conclusion, repeated month after month in 2026, is clear: where AI is used most, there is neither higher unemployment nor lower employment, and the types of jobs offered by the market are not changing in a way that can be attributed to AI either. What is observed is stability.

In Europe, the European Central Bank — based on its SAFE survey of around 5,000 euro area firms — found that companies using AI intensively are not replacing jobs and that, on average, they are somewhat more likely to hire additional staff. That said, the ECB itself qualifies this by noting that the positive net effect is driven by companies using AI for R&D and innovation. Those adopting it explicitly to cut labour costs do show lower hiring and more layoffs. The intention behind adopting the tool matters.

On the productivity side, a European Investment Bank working paper (EIB, 2026/02), also summarised in VoxEU/CEPR by authors from the EIB and the BIS, analysed more than 12,000 non-financial firms in the EU and the US and estimated that AI adoption raises labour productivity by around 4% in the EU, driven by capital deepening and not by workforce reductions, and that there is no evidence of job losses in the short term. The benefit, however, is unevenly distributed and concentrated among medium-sized and large firms, which are the ones with the capacity to invest in data, software and human capital.

Where demand is growing

If AI were hollowing out the labour market, we would not expect to see hiring linked to it grow — which is precisely what is happening.

PwC’s Global AI Jobs Barometer 2026, based on more than one billion job ads across 27 countries, describes a two-sided market: jobs in which AI automates routine tasks and brings judgement, creativity and leadership to the fore are growing faster and also seeing stronger wage increases than the rest. Jobs requiring specific AI capabilities grew by 69%, compared with 9% for the overall market.

Indeed’s Hiring Lab observed the same pattern in the US, while warning that this growth in job postings mentioning AI is taking place in a weak hiring market. AI is not inflating employment; it is simply concentrating the available hiring around certain roles and skills.

That AI is complementing workers rather than replacing them can also be seen in the Anthropic Economic Index, which shows that augmentation — that is, cases where the user learns, iterates or validates with AI — has surpassed pure automation. A 2026 academic preprint, The AI Skills Shift, quantifies the same idea, estimating that 78.7% of observed interactions with AI are augmentation rather than automation. And another study on 35 European countries, From Exposure to Adoption, still finds no detectable effect of early adoption on the task restructuring reported by workers themselves, suggesting that we are still in a phase of integration, not transformation.

But “yet” does not mean “never”

But it would be dangerous to close on a triumphalist note. The same reports that defuse the immediate alarmism also point to where the pressure is coming from.

The most explicit is the BCG Henderson Institute report whose message — “AI will reshape more jobs than it replaces” — estimates that just over 50% of jobs in the US will be reshaped within 2-3 years and that around 15% could disappear within 5 years. The fact that job destruction is not visible today does not eliminate that possibility in the medium term.

There are also two specific risks on which there is broad agreement:

Conclusion

For a company, the message should be neither complacency nor panic. It is that the advantage does not go to whoever has better tools, but to whoever reconfigures their processes in order to use them. AI is not complementary to workers just because it exists; it requires deliberate and conscious intention and investment. That is why the benefits are concentrated, for now, in companies that can afford it and that know very well what they are doing.

Therefore, a balanced conclusion avoids both alarmism and denial. And what it does require is updating skills, rethinking roles and designing transition policies before the medium-term curve becomes visible.

Sources