Task crossover

AI does not only transform how work is performed. It also redraws boundaries between occupations.

In an analysis of more than 800,000 messages from US ChatGPT users, OpenAI Economic Research found that 43.5% of non-generic, occupation-specific messages concerned a task historically associated with another occupation. The report calls this phenomenon ‘task crossover’. This figure concerns analysed uses, not 43.5% of jobs.

A salesperson can now explore a customer dataset that would previously have been sent to an analyst. A marketing professional can troubleshoot a website or write a simple script without waiting for a developer. AI therefore changes not only how work is carried out, but also how it is distributed.

At first sight this may appear positive: greater autonomy, faster responses and broader skills. But an accessible task is not necessarily a task that has been mastered, recognised or made sustainable.

A reading through the Gollac factors

The framework proposed by the Gollac report helps examine this reconfiguration without prematurely concluding that it is either progress or a risk.

  • Work intensity — New assignments may be added without reducing the original workload, deadlines or objectives.
  • Autonomy — A worker may be able to do more while also having to take responsibility for tasks they do not fully master.
  • Social relations — Using AI may reduce exchanges with specialist colleagues or shift verification work to other teams.
  • Conflicts of values — People may have to produce work quickly without being able to assess it with sufficient certainty.
  • Job insecurity — Occupational boundaries, expected skills and the value assigned to expertise may become less certain.

Observe the transformation before it becomes invisible

The figure of 43.5% does not by itself demonstrate a psychosocial risk or a health effect. It describes uses of ChatGPT in the United States and relies on an occupational and task classification. It is nevertheless a signal: the real content of jobs may begin to change before job descriptions, organisations or labour statistics do.

This is precisely why occupational-health professionals should engage with the issue quickly. Together with employers, HR teams and Social and Economic Committees (CSEs), they can document tasks entering an occupation, tasks disappearing, time genuinely saved, training needs, responsibility for verification and effects on teams.

Models, and the systems that incorporate them, should therefore be assessed before deployment—not only for technical performance, but also for their possible effects on work and worker health.

The question may not only be whether AI broadens what a worker can do, but whether the organisation genuinely gives that worker the means, time and recognition needed to do it well.

Point to watch

An accessible task is not automatically an acquired skill. We must distinguish the ability to produce an output with AI from the ability to verify its quality, take responsibility for its consequences and integrate it sustainably into an occupation.