AI was supposed to hit young office workers. The data does not back that up

Robert Fairlie and Jane Wu, two researchers at CESifo in Munich, published a paper called "The Early Impacts of AI on Employment Among Recent College Graduates." They write plainly that there is "no evidence of significant, widespread displacement or reduced hiring of college graduates, either in absolute or relative terms."

They picked that group for a reason. Shifts in labor demand should show up first at the entry point, in fewer hires for starting positions. The paper runs against an earlier Stanford study, which found that entry-level employment in AI-exposed occupations is lagging behind other fields. Ars Technica notes, citing the authors, that no data so far describes mass displacement.

That matters because the narrative and the hard data have drifted apart in time. Deployments keep moving ahead, mostly under the banner of support rather than replacement. Adecco Group is rolling out Agentforce Coworker, an assistant built on Anthropic's Claude model, to 27,000 employees in more than 40 countries, after pilots in the United Kingdom and France. The company signed an agreement with Salesforce for unlimited access to the Agentforce 360 platform, so it can scale agentic tools across the group.

Adecco already uses AI agents in recruitment in ten countries that account for half of the group's business revenue. The new tool, integrated with Salesforce, is meant to consolidate data and knowledge previously scattered across dozens of systems. It will also draw on context from more than 2.5 million interactions between AI agents and candidates recorded since April 2025.

The scope of use is very office-bound. In sales teams, the assistant is to find priority clients, prepare notes, fill in data and check lead status across teams. Recruiters are to select candidates and launch agents that support pre-screening, hiring and onboarding of new people.

Both pictures are true at the same time. What gets automated is the execution of a task, not the position. Whether that translates into less entry-level hiring depends on whether companies prefer to handle the same workload with a smaller team. For now the statistics do not confirm the worst-case scenario. That does not mean they will not confirm it a year from now.

Data on deployments shows the other side of the phenomenon. When a large services company gives an AI assistant to tens of thousands of employees in dozens of countries, it changes not only productivity but also the mix of tasks. A recruiter spends less time tidying data and more time talking to candidates. The employment effect of such a rollout therefore depends on whether the company decides to handle a larger workload with the same team, or simply with a smaller one. The statistics so far do not settle that question. They measure employment levels, not what newly hired people actually do.