AI and Jobs in 2026: What the Labor Market Data Really Shows

Three years after ChatGPT, the AI-and-jobs debate finally has data. It points in two directions at once. In the US, AI has become the most-cited reason for announced layoffs in 2026, and young graduates in AI-exposed occupations are being hired far less often. Yet total unemployment has barely moved, most workers report that AI makes them faster rather than redundant, and the AI labs' own usage data shows Claude and ChatGPT augmenting work about as often as automating it. This article brings together the most important numbers from Anthropic, OpenAI, Stanford, the ILO, Challenger and Eurostat, and explains what they mean for employers in Croatia and the DACH region.

The short answer: no mass unemployment, but a closing door at the entry level

If you only remember one thing: the impact of AI on jobs in 2026 is concentrated, not general. Stanford economists Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, who analyse ADP payroll records covering millions of US workers, find "no evidence of widespread, economy-wide job displacement." What they do find is a widening gap for the youngest workers in occupations where AI can take over tasks, and it shows up as fewer hires, not more layoffs. Companies are not firing their accountants and developers; they are hiring fewer juniors to replace those who leave.

Layoffs: AI is now the #1 stated reason in the US

Outplacement firm Challenger, Gray & Christmas records the reason employers give for every announced US job cut. In 2025, AI was cited for 54,836 cuts. In 2026 it overtook every other reason: by May it had already been named in 87,714 announcements, and May alone (38,579) was the highest monthly total since tracking began in 2023. Through September the count reached 120,136, about 21% of all announced cuts and still the leading reason year-to-date. Technology companies account for 29% of all 2026 cuts.

Young workers are the canaries in the coal mine

The Stanford team first reported in August 2025 that workers aged 22 to 25 in the most AI-exposed occupations, such as software development and customer service, had seen a 13% relative decline in employment since late 2022. Their revised paper of 12 August 2026, with data through June 2026, puts the gap at 19% below where employment would be had it kept pace with less-exposed peers. Three details matter for employers:

• It is a hiring effect. The gap comes mainly from fewer young people being hired, not from more of them being let go. The Dallas Fed reached the same conclusion: the drop is in people moving straight from education into a first job.
• Substitution versus complement. Declines cluster where AI replaces tasks. Where AI complements workers, employment is flat or rising, especially for experienced staff.
• The effect survives the obvious objections. It persists after excluding tech firms, and after controlling for interest rates and remote work, although it shrinks when controlling for education.

The macro effect is still small: the Dallas Fed estimates that even if all of this missing youth employment had turned into unemployment, it would explain only about 0.1 percentage points of US unemployment. But the pipeline problem is real. US employers project only a 1.6% hiring increase for the class of 2026 (NACE), and a quarter of them say AI has already reduced the need for tasks entry-level staff used to do.

What the AI labs' own data says

Anthropic: augmentation still edges out automation

Anthropic's Economic Index classifies millions of anonymised Claude conversations by how people use the model. In augmentation, a person iterates with Claude, learns or checks its work. In automation, they hand over a task with little input. On Claude.ai the two have been neck and neck: automation briefly led in August 2025 (49% vs. 47%), but by November 2025 augmentation was back ahead at 52% vs. 45%, which Anthropic links partly to features such as file creation, memory and Skills. Business use is a different story: about 75% of enterprise API traffic is automation, a sign that companies are wiring Claude into back-office workflows.

In June 2026 Anthropic went further and linked usage data to a survey of around 9,700 Claude users. The results are more optimistic than the headlines: 86% say AI makes them faster, 82% that it widens the scope of what they can do and 57% that it has made their skills more valuable. Only 10% rate their own job loss as likely, slightly below the normal US involuntary separation rate, and just 38% of those attribute the risk to AI. But the same respondents are worried for others: more than a third put the job-loss probability of a junior colleague above 60%, and over a third expect AI to handle most of their tasks within 12 months. Workers with 15+ years of experience report about 10 points less AI task share than first-year workers, which matches the Stanford pattern.

OpenAI: measuring whether models can do the job

OpenAI approaches the question from the capability side. Its GDPval benchmark asks industry experts with an average of 14 years of experience to blindly compare AI deliverables with human work across 44 occupations in the nine sectors that contribute most to US GDP, from legal briefs to engineering drawings. In the first edition (September 2025), Claude Opus 4.1 won or tied against experts in 47.6% of tasks and GPT-5 in 38.8%, and newer models have kept climbing. OpenAI also launched a jobs platform and certification programme to help workers prove AI skills. GDPval's limit is important, though: it measures well-specified deliverables, not the meetings, judgement calls and client relationships that make up much of a real job.

Mistral and Manus: two ends of the spectrum

Manus, now owned by Meta, represents the fully delegated end: autonomous agents that research, build and execute multi-step work with minimal supervision, exactly the "automation" pattern in Anthropic's data. Mistral represents a European option for the other end: open-weight models and Le Chat Enterprise that companies can run under EU control and adapt to their own processes. For HR and works councils in Europe, that difference in control often matters as much as raw capability.

Exposure is not job loss

Global estimates confirm the picture. The ILO's refined index (May 2025) finds that about one in four jobs worldwide is exposed to generative AI, rising to 34% in high-income countries. But only 3.3% of global employment sits in the highest-exposure category, where most tasks could be automated, and women are overrepresented there (4.7% vs. 2.4% of men) because of clerical work. The ILO's conclusion: transformation is far more likely than replacement. The World Economic Forum's Future of Jobs Report 2025 points the same way, expecting 170 million jobs to be created and 92 million displaced by 2030, a net gain that still requires huge reskilling.

Croatia and the DACH region: the adoption gap

In Europe the bigger story in 2026 is still adoption, not displacement. According to Eurostat, 19.95% of EU enterprises with at least 10 employees used an AI technology in 2025, up from 8% in 2023. Austria (29.95%) and Germany (25.97%) are well above average. Croatia has nearly doubled since 2023 but, at 15.19%, is still behind the EU and its neighbour Slovenia (21.61%).

For Croatian companies, low adoption is a double-edged sword. It means the US entry-level squeeze arrives later, but it also means lost productivity at a time when labour shortages and ageing workforces already limit growth in Croatia, Germany and Austria. In that context, AI that augments existing staff is less a threat than a way to keep growing without hiring people who are not there. Our AI adoption report and guide for SMEs show where to start.

What employers should do now

• Audit tasks, not jobs. Break roles into tasks and mark which ones AI can automate, which it can speed up and which need a human. That is where both savings and risks become visible.
• Augment first, automate second. Start with use cases where staff work with Claude or ChatGPT and stay accountable, then automate stable, well-defined back-office flows via the API. Measure results with a clear ROI framework.
• Protect the junior pipeline. If AI does the work juniors used to learn on, redesign entry-level roles around reviewing AI output, customer contact and supervised projects. Today's missing juniors are tomorrow's missing seniors.
• Invest in AI literacy. Since February 2025 the EU AI Act (Article 4) requires companies that deploy AI to ensure sufficient AI literacy among their staff. See our guide to workforce upskilling.
• Involve employee representatives early. In Germany and Austria, works councils often have co-determination rights when software can monitor performance, and AI used in hiring or worker management is classified as high-risk under the AI Act. See AI in HR.
• Communicate honestly. The surveys show people fear for colleagues more than for themselves. A clear plan for which tasks change, and how people will be trained, reduces resistance and shadow AI.

The bottom line

The 2026 data rejects both extremes. AI is not yet causing mass unemployment, but it is no longer neutral either: it has become the leading stated reason for US layoffs and is quietly closing the entry door for young workers in exposed occupations. At the same time, most people who use Claude and ChatGPT every day say it makes their work faster, broader and more valuable. For European companies, the risk is less that AI takes jobs tomorrow and more that competitors use it to do more with the people they have. The winners will be those who redesign work deliberately, keep training juniors and treat AI as a colleague, not a cost line.