Everyone agrees that artificial intelligence (AI) is changing the economy. What almost nobody has been able to show, until now, is how ordinary people actually use it day to day. Google has just tried to answer that question with the largest study of its kind, and the findings are surprising, encouraging, and a little humbling all at once.
What is ATLAS?
On July 23, 2026, Google published the first edition of its AI & Economy ATLAS report. ATLAS stands for Activity, Task, Landscape, and Adoption Study. In plain English, it is an ongoing effort to measure what people are doing when they talk to Google’s AI tools.
The scale is enormous. The report is built from 15 million anonymized human-AI conversations across products such as the Gemini app, AI Mode in Search, and the Gemini API; tools used by more than a billion people every month. The data covers over 150 countries, 140 languages, 800 occupations, and 4,000 different tasks. Google calls it “the most comprehensive look to date at how real people are using AI at scale.”
A quick note on privacy: “De-identified” and “anonymized” mean the data has been stripped of anything that could point back to an individual person. Google says it removed personal details, cut the links between the study data and users’ original activity, and grouped everything into summaries covering many people at once.
Five things the study found
1. AI use at work is “broad but shallow.” AI has spread into every industry sector and into 68% of all occupations (jobs that make up 90% of U.S. employment). But within any given job, people only use it for about 21% of their tasks. In other words, lots of people are dipping a toe in, but few are diving all the way in.
2. AI is mostly a helper, not a replacement. This is perhaps the most important finding for anyone worried about their livelihood. Most workplace AI use is collaborative — brainstorming, planning, looking things up, and learning. Google found that fewer than 10% of work interactions fully automated a task. The machine is acting more like a colleague than a substitute.
3. It’s not just office workers. AI isn’t only for so-called “knowledge workers.” Auto technicians, industrial mechanics, and other hands-on tradespeople are using conversational AI as a real-time assistant to diagnose problems, debug wiring, and inspect machinery. These workers were twice as likely to use multimodal AI, meaning AI that works with images and video, not just text.
4. Most AI use happens at home, not at work. A striking 86% of interactions took place outside the workplace. People use AI to research purchases, figure out how to use appliances, and, notably, wrestle with frustrating admin tasks like taxes, licensing, and fines. Google’s point here is subtle but important: this kind of value often doesn’t show up in official economic statistics like GDP, so the real benefit of AI may be larger than the numbers suggest.
5. Richer countries use AI more — but not always. AI has reached countries representing 99% of the world’s population, and English makes up only about a third of conversations. However, usage per person tends to track a country’s wealth, which raises concerns about a “digital divide.” The bright spot: some middle-income nations in South America and the Middle East are adopting AI as fast as wealthier countries.
Key concepts, explained
To follow the debate around this report, a few terms are worth knowing:
Automation vs. augmentation. Automation means a machine does a task instead of a person. Augmentation means the machine helps a person do the task better or faster. ATLAS suggests today’s AI leans heavily toward augmentation — a distinction that MIT labor economist David Autor, who contributed to the report, has long emphasized in his work on how technology reshapes rather than simply eliminates jobs (MIT Economics; VoxDev, “David Autor on AI and the future of work,” Nov 2025).
Tasks vs. jobs. Economists increasingly analyze work as a bundle of tasks rather than a single job. AI might handle a few of your tasks while leaving the rest to you — which is exactly the “broad but shallow” pattern ATLAS describes.
Non-routine cognitive work. This means creative or judgment-heavy thinking, like design or testing ideas, as opposed to repetitive work. These tasks showed up in AI conversations far more often (65%) than they appear in the economy overall (35%).
GDP and “missing” value. GDP measures the value of goods and services bought and sold. When AI helps you file your taxes at home for free, that benefit is real but largely invisible to GDP; an example of what economists call unmeasured or non-market value.
How Google built it
The insights come from a Google DeepMind tool called OCTO (Observation Clustering and Taxonomy Organisation), which takes huge amounts of messy conversation text and sorts it into organized categories. Google also credits Dame Diane Coyle of Cambridge and Dr. David Autor of MIT for their contributions.
What it means and what comes next
For businesses and workers, the reassuring headline is that AI, at least for now, is behaving like a capable assistant rather than a replacement. The bigger opportunity may lie in the tasks people find tedious, both at work and at home.
Google is careful to call this “an early view of a quickly moving landscape,” and notes that ATLAS doesn’t capture everything, it leaves out heavily used products like Google Workspace, Google Translate, and enterprise tools. As the technology advances and the study grows, the picture will keep changing.
For anyone trying to understand where the AI economy is heading, the message from ATLAS v1.0 is clear: adoption is wide, the technology is helping more than it’s replacing, and much of its value is quietly showing up in our everyday lives.
Suggested further reading / citations:
- Google, “Understanding the AI economy” (ATLAS v1.0 report), July 23, 2026
- MIT Economics, David H. Autor faculty page
- VoxDev, “David Autor on AI and the future of work”, Nov 2025
- Issues in Science and Technology, “David Autor: How Is AI Shaping the Future of Work?”, Jan 2026
