Tech pundits have spent months predicting that generative AI will soon automate vast swaths of white-collar jobs. But a new analysis from Google Research, based on 15 million real interactions with its Gemini platform, suggests that reality is far less dramatic. The paper, released last week, introduces the AI & Economy ATLAS—a study of how workers actually prompt Gemini in the app, AI Mode, and API. The researchers’ blunt conclusion: “We do not find evidence… to support the claims that AI is about to cause massive automation and displacement of white-collar work.”

The study classified work-related prompts using a combination of U.S. Bureau of Labor Statistics occupational codes and O*NET task descriptions. While the classification required some probabilistic estimation for ambiguous interactions, human reviewers confirmed the approach was reliable. The data shows that AI usage across occupations remains “shallow and overwhelmingly collaborative,” with end-to-end task automation appearing only in limited cases. Essentially, workers are using Gemini as a co-pilot—getting drafts, summaries, or suggestions—not as a replacement for their own judgment.

Not surprisingly, occupations overrepresented in Gemini’s usage logs included software developers, financial and market analysts, and systems administrators—roles that involve heavy information processing. Conversely, salespeople, transportation workers, and food preparation staff were among the least likely to use the tool for on-the-job tasks. For data and ML engineers, the takeaway is twofold: current AI adoption patterns don’t signal a workforce apocalypse, and the real opportunity lies in building tools that augment human decision-making rather than trying to automate entire workflows end-to-end.

Source: Ars Technica