Fresh perspectives from Google's AI & Economy ATLAS

New findings from Anthropic's Economic Index... wait, actually these results come from ATLAS, a long-term research initiative tracking how AI is reshaping work across the globe. The latest release maps AI adoption by occupation, income level, and region, and pairs it with a joint study from Google, Google DeepMind, and MIT FutureTech examining how scientists put AI to work. Together, the two pieces of research offer one of the clearest pictures yet of where AI is actually being used — and where its promised productivity gains are still waiting to materialize.

Which Occupations Lead AI Adoption

The ATLAS data shows a clear split between OECD and non-OECD countries. In OECD nations, computer and mathematical occupations along with business and financial operations top the list of AI users. In non-OECD countries, the picture looks different: office and administrative support roles lead, followed by arts, design, entertainment, sports, and media occupations, as well as educational instruction and library work.

Income levels matter, too. AI adoption generally rises with a country's income, but some economies outperform what their GDP per capita alone would suggest. Brazil and the UAE both show adoption rates well above expectations for their income brackets.

Even hands-on work is seeing AI involvement. Real-time equipment diagnostics and troubleshooting powered by AI vary noticeably by region: in Brazil and Germany, 7% of work-related AI usage goes toward manual tasks — 1.4 times the global average — while Japan sits at just 4%.

Scientists Are Ahead of the Curve

A new study from Google, Google DeepMind, and MIT FutureTech, built on ATLAS data, examines AI use in scientific research. The analysis covers 2,600 specialized AI models and draws on a survey of more than 600 scientists in the U.S. and U.K., all organized through a new taxonomy from MIT FutureTech that maps what scientists actually do.

The headline finding: scientists use AI at higher rates than most other occupations, and nearly half use some form of AI every day. But they rely on different tools for different jobs. General-purpose LLMs like Gemini see broad use across scientific fields and task types, while specialized AI models cluster in health and life sciences and in domain-specific data prediction, generation, and simulation.

Time Saved, Bottlenecks Gained

The productivity story is mixed. Scientists report saving just under seven hours per week thanks to AI — time that can flow back into research. Yet those gains don't automatically become new discoveries.

The study finds scientists spending considerable time validating AI outputs, an accumulating backlog of untested hypotheses, and new chokepoints in physical experimentation and clinical validation. In other words, science faces the same challenge as most occupations: AI's potential to boost productivity is real, but realizing it at scale may require rethinking scientific processes and workflows to keep pace with rapidly evolving AI capabilities.

What Comes Next

Plenty of open questions remain about AI's economic impact. ATLAS is designed as a long-term research project, and its team plans to work with academic partners and others to identify new research directions — aiming to deepen understanding of how AI is transforming the economy as the technology continues to evolve.

Meta description: ATLAS data reveals AI adoption by profession and region, plus a new study showing scientists save nearly seven hours weekly using AI tools.

Tags: AI adoption, ATLAS research, scientists using AI, productivity, OECD countries

Featured image: Abstract visualization of interconnected data nodes and flowing lines across a world map, rendered in cool blue and violet tones.