Does AI adoption follow wealth and mature over time?

Anthropic's September 2025 Economic Index report, drawn from a sample of Claude.ai conversations and Anthropic's own 1P API traffic, finds that Claude usage is "geographically concentrated" and tracks national income. Singapore and Canada use Claude at 4.6x and 2.9x their population share, while Indonesia (0.36x), India (0.27x) and Nigeria (0.2x) trail; across countries, "a 1% increase in GDP per capita [is] associated with a 0.7% increase in Claude usage per capita." The unevenness also shows up in what adoption looks like: "Lower-adoption countries tend to see more coding usage, while high-adoption regions show diverse applications across education, science, and business" — coding is "over half of all usage in India" against "roughly a third of all usage globally."

The report's own explanation is a maturity curve rather than a fixed national trait: "After controlling for task mix by country, low AUI countries are more likely to delegate complete tasks (automation), while high-adoption areas tend toward greater learning and human-AI iteration (augmentation)." The same split separates channels within Claude's own product line — API usage runs "automation dominant" (77% of business uses show automation patterns, versus "about 50% for Claude.ai users," and 97% of economic tasks versus 47% on Claude.ai), which the report attributes to "the programmatic nature of API usage." It also ties sophisticated deployment to information access rather than raw capability: "curating the right context for models will be important for high-impact deployments of AI in complex domains," and "costly data modernization and organizational investments to elicit contextual information may be a bottleneck for AI adoption."

This sits beside Where have workers actually delegated tasks to AI? and Is AI creating common skills across jobs or deepening divisions?, which locate concentration in task type and skill demand; this report adds a geographic and lifecycle axis — concentration isn't only which tasks get automated, but which countries and channels sit further along the path from automation to augmentation. It also complicates Does AI assistance erode the skills needed to oversee it?: those engineers work inside a high-income, augmentation-leaning market by this report's own framing, so their low delegation ceiling may describe where the US sits on the curve rather than a universal limit on delegability. And it gives a geographic reading to How are national lab staff actually using generative AI? — a lab whose use is "largely experimental" and copilot-mode looks like what this report would predict for an organization early on the adoption curve, regardless of national income.

The report does not establish that the automation-to-augmentation shift is causal: it frames the country-level pattern as correlational, "perhaps reflecting differences in how AI is deployed by economies at different stages of structural transformation," and the task-mix control is Anthropic's own method, not independently audited. The usage data describes Claude specifically, drawn from Anthropic's own traffic, not AI adoption generally, and a vendor measuring adoption of its own product has a stake in the story that usage matures rather than plateaus. If the pattern holds, today's coding-heavy, automation-heavy usage in lower-income countries would be a stage rather than a ceiling — but the report commits only to tracking "whether these adoption gaps narrow, widen, or change in structure over time," not to which way they will move.