ANTHROPIC
Some countries are swimming in AI, while others barely touch it
AI is spreading far faster than past tech.
Around 40% of US employees now use AI at work, double since 2023.
It rides today’s digital rails and is easy to try, so early use clusters in a few regions and task types before widening.
Anthropic tracks this via Claude.ai (consumer) and API (enterprise) data.
Coding still leads, but education and science are growing.
New features (web search, research modes, Artifacts) nudged more internet research and multimedia/app creation.
People are delegating more end-to-end work: “do the whole task” requests rose, and usage shifted from debugging to creating code, pointing to better first-pass quality and rising confidence.
Where AI is used and how:
Per-capita usage concentrates in richer, tech-ready economies.
Many emerging economies are below what their population would suggest.
In the US, DC leads per-capita, then Utah and California. Local economies shape usage: California skews to IT, Florida to finance, DC to documents/careers.
Lower-adoption places lean heavily on coding; higher-adoption places spread across education, science and business.
Collaboration shifts too: lower-adoption areas automate more, higher-adoption areas iterate and learn more, even after adjusting for task mix.
Programmatic use looks more specialised and more automated.
Software development and office/admin tasks dominate; education/creative are smaller than on Claude.ai.
About three-quarters of API traffic is full task delegation.
Capability beats price in driving use, but complex tasks need rich context, so data pipelines and process changes become the bottleneck.
Collab vs command
This is the classic early phase of a general-purpose tech: narrow, automation-heavy, productivity-raising.
Benefits may concentrate in already-advantaged regions and roles.
Routine work could be displaced, while people with organisational know-how, and who can steer AI, gain most.
Anthropic has open-sourced task-level usage for Claude.ai and first-party API (plus methods).
Geographic cuts currently cover Claude.ai, to help researchers test what drives adoption, where gains accrue, and which workers are most affected.
Enterprises said: collaboration? Never heard of her. Automation only.