AI turns people into generalists: Almost half of messages ask about tasks from other professions

People are crossing the boundaries of their roles

OpenAI's economic research team analyzed over 800,000 work-related messages from US ChatGPT users. They found that 43.5% of occupation-specific messages deal with tasks historically belonging to a completely different profession.

This phenomenon, which OpenAI calls 'task crossover,' means that people are not primarily using AI just to do their current jobs faster. Instead, they use it to expand their scope and complete assignments that would previously have required a colleague from another department.

Smaller teams benefit the most

The analysis shows that some fields act as task 'exporters.' This applies mainly to marketing and engineering. These tasks are often taken on by people from other departments thanks to AI. Conversely, people in design, sales, or HR most frequently take on tasks that previously didn't fall under them.

Moreover, this effect is much stronger in smaller companies. Where there are fewer resources to hire specialists, ChatGPT serves as a tool that allows a smaller team to cover a wider range of activities. In short, people are learning to do work with the help of AI that they otherwise wouldn't have the budget for.

Primary data replaces surveys

The limit of this research lies in the data itself. OpenAI observes what people ask the model, but it cannot see whether the result is actually usable in production, or if it was just a futile attempt.

Even so, it is an important step away from surveys like 'how many people fear for their jobs.' Tracking actual ChatGPT logs shows real behavior, not just declared intentions from questionnaires.

Generating code by marketers will decide

It's worth watching whether this trend persists and translates into demands on the labor market. The proof will be when companies stop looking for narrowly profiled specialists and start rewarding the ability to solve complex cross-departmental tasks with the help of AI more.

Lilith's verdict

The question is no longer how to make a marketer 10% faster, but whether that marketer can spin up a database by themselves. Corporate silos will crack as soon as someone sends enough cheap tokens their way.