Anthropic has published a new dataset, the Anthropic Economic Index, that tracks how its AI assistant Claude is being used across the world. Rather than relying on anecdotes or surveys, the index draws on aggregated usage data to chart adoption by country, by US state, and by occupation.

At the centre of the dataset is what Anthropic calls a Usage Index. This figure compares actual AI usage in a region against what would be expected based purely on population size. A score of 1.0 means usage matches population share exactly. Anything above 1.0 indicates usage outpacing population, and anything below suggests the opposite. It is, in effect, a per-capita measure of how embedded AI tools have become in daily working life, stripped of the distortion that comes from simply counting raw numbers in large countries.

The picture this produces moves the conversation away from general claims about AI's potential and towards a more concrete question: where is this technology actually being put to work, and for what.

Globally, Australia sits at the top of the table, with a Usage Index of 6.40, meaning Australians are using Claude at more than six times the rate their population alone would predict. Singapore follows at 5.81, and Switzerland at 5.02. These three economies share certain features. Each has a highly digitised public and private sector, a disproportionately large finance or professional services industry, and English-language fluency that lowers the barrier to adopting an English-trained AI model. By contrast, the regions further down the global map, many in parts of Africa, South Asia and Latin America, show usage closer to or below their population share, suggesting adoption is still in its early stages rather than absent altogether.

Within the United States, the pattern repeats at a smaller scale. The West Coast and the Northeast corridor dominate the "leading" category on Anthropic's state-by-state map, areas long associated with technology, finance and higher education.

One region stands apart from the rest. Washington, D.C., ranks first out of the 51 areas measured, with a Usage Index of 3.32. That figure dwarfs the scores recorded in California (1.62) and New York (1.55), states more commonly associated with the technology sector. The gap is striking enough to warrant closer examination of what, specifically, people in the capital are using AI for.

The answer lies in the index's task-level breakdown. In Washington, the most common use case is self-presentation writing, accounting for 5.8% of recorded tasks, a category that likely includes CVs, cover letters and professional biographies. Homework follows at 5.7%, and general workplace writing at 3.6%. These are not requests for AI to run entire projects unsupervised. They are requests to help draft, refine or accelerate work that a person is still ultimately producing themselves.

The occupational data reinforces this. The Computer and Mathematical sector accounts for the largest single share of usage in D.C., at 18.5%, which is unsurprising given the technical nature of much government and contractor work in the city. Less expected is the scale of adoption in Arts, Design, Entertainment, Sports and Media, at 16.4%, and in Educational Instruction, at 13.7%. Together these figures suggest AI use in the capital extends well beyond software engineers and policy analysts, reaching into communications, teaching and creative roles that depend heavily on writing and presentation.

Taken as a whole, the index points more towards augmentation than automation. The dominant tasks, writing, editing, research support, homework assistance, are ones where a human remains in control of the final output, using AI to speed up a draft or sharpen a sentence rather than to replace the work entirely. Whether that balance shifts as the technology matures is a question the next edition of the index may begin to answer.