Google study finds millions use AI at work, but full job automation remains limited

250

Artificial intelligence is spreading across occupations, but workers are still using it mainly as an assistant rather than handing over entire jobs. Google’s new ATLAS study examined 14.65 million de-identified interactions with the Gemini app, AI Mode and Gemini API during two weeks in April, finding that research, drafting, troubleshooting, and learning dominated usage.

The finding matters for India because companies face pressure to deploy generative AI while proving measurable productivity gains. The evidence suggests that the immediate business case lies in redesigning individual tasks, not treating AI as a justification for indiscriminate job cuts.

AI adoption is also extending beyond conventional technology roles. The study found unexpectedly strong use among electricians seeking wiring diagrams and automotive workers looking for engine information. Yet adoption remained higher among well-paid professionals and wealthier regions, raising the risk that unequal access and AI literacy could widen existing productivity gaps.

India combines substantial demand with considerable untapped potential. A Google-Kantar study of more than 8,000 people across 18 Indian cities found that only 31% had tried generative AI, while 75% wanted a “daily growth collaborator.” Among Gemini users, 93% reported improved productivity and task management.

Google described Indians as having “a strong, innate desire…to improve and excel.” The phrase captures the commercial opportunity, but self-reported gains are not the same as independently measured output.

There is another limitation: ATLAS excluded workplace products such as Gemini Enterprise and Workspace, meaning it may understate business automation.

Indian employers should therefore measure time saved, error rates and customer outcomes before restructuring roles. AI may be the co-pilot today, but management still decides the destination—and remains accountable when the navigation fails.