Google's ATLAS study: AI reaches 68% of jobs but rarely automates

The story that has bounced through boardrooms for two years is that AI is about to hollow out white collar work. Google just published a data set that pushes back on the timing, if not the direction, of that story.

According to reporting in Ars Technica, Google's first AI & Economy ATLAS report analyzed 15 million de-identified Gemini interactions across 150 countries, 140 languages, 800 occupations and 4,000 tasks. The headline finding is that adoption is broad but shallow. AI is already being used in 68% of occupations, representing around 90% of total US employment, but within a typical job it only touches about 21% of tasks. Only 3% of occupations show AI used for over 75% of their tasks.

The part that undercuts the replacement narrative is how workers are actually using the models. Google reports that less than 10% of those interactions fully automate a task. The vast majority are collaborative uses like ideation, strategy, information retrieval, and learning. People are using Gemini as a partner on a slice of what they do, not handing it a job and walking away.

The caveats matter. This is a v1.0 report from a company selling the tool being measured, so the collaboration framing lands where you would expect. It's also Gemini-only. Anthropic's Economic Index, launched in February 2025, reported that roughly 36% of occupations were using AI for at least a quarter of their tasks, a directionally similar but noticeably heavier picture. Aggregate averages across 800 occupations also hide the roles where displacement may be real, and the reporting doesn't spell out how ATLAS separates automation from collaboration.

For leaders pricing in AI-driven headcount cuts over the next two budget cycles, the useful signal here is calibration. If the largest public dataset so far shows full task automation in single digits, the productivity gain is more likely to arrive as workers doing more of the same job than as roles quietly disappearing. That is a different investment plan.