Understanding the AI economy

Source: Google Date: 2026-07-23 URL: https://blog.google/innovation-and-ai/technology/research/understanding-the-ai-economy/

Summary

Google published ATLAS, a study of 15 million de-identified Gemini interactions across 150+ countries and 140 languages. Headline finding: AI reaches 68% of occupations (90% of U.S. employment) but is applied to only ~21% of tasks within a given job, and under 10% of interactions fully automate a task — the dominant pattern is ideation, strategy, and troubleshooting assistance, not replacement. Usage also shows up heavily outside white-collar knowledge work (manual trades diagnostics) and outside employment entirely (86%+ of interactions are non-work), with adoption tracking GDP per capita globally.

Implications

• Work-AI-adoption thread: this is a large-scale empirical counterweight to “automation is imminent” narratives — the data shows augmentation (ideation/strategy) dominating over task automation by a wide margin, which should anchor any adoption-timeline projections against overclaiming near-term labor displacement.
• Open≠local / capability clocks: adoption breadth (68% of occupations touched) versus depth (21% of tasks) is a useful two-axis frame for tracking how fast frontier capability actually converts into workplace practice, independent of how fast model benchmarks move.