The latest Anthropic Economic Index highlights the rapid yet uneven adoption of artificial intelligence across regions and industries. In the United States, 40% of employees now report using AI at work, up from 20% in 2023, reflecting a swift integration into daily workflows. This pace surpasses earlier technologies like the internet, which took five years to reach adoption levels that AI achieved in just two. The speed is driven by AI’s ease of use, compatibility with existing digital systems, and continuous improvements in model performance.
Historically, transformative technologies such as electricity and personal computers required decades to become widespread. In contrast, AI is diffusing faster but remains concentrated in specific geographic areas and enterprise functions. Early adoption patterns typically favor certain regions and narrow applications before expanding more broadly—a trend clearly visible in current AI usage.
To better understand these dynamics, the index expands its analysis to include geographic data from over 150 countries and all U.S. states, along with new insights into enterprise API usage. These additions reveal how AI tools like Claude are being used differently depending on location and organizational context.
Over the past eight months, usage on Claude.ai has evolved significantly. Tasks related to education have grown from 9.3% to 12.4% of total interactions, while scientific applications increased from 6.3% to 7.2%. Coding remains dominant at 36%, but there’s a noticeable shift toward creating new code (+4.5 percentage points) rather than debugging (-2.9 percentage points), suggesting users are accomplishing more in single sessions.
User behavior also shows a growing preference for delegating full tasks to AI. Directive-style conversations—where users assign complete tasks with minimal back-and-forth—rose from 27% to 39%. This indicates increasing confidence in AI outputs and possibly reflects improved model reliability.
Geographically, AI usage correlates strongly with national income levels. Singapore and Canada lead in per capita usage, operating at 4.6x and 2.9x their expected levels based on population size. Conversely, emerging economies like Indonesia (0.36x), India (0.27x), and Nigeria (0.2x) show significantly lower engagement. To measure this disparity, the report introduces the Anthropic AI Usage Index (AUI), which compares actual usage to expected usage relative to working-age populations.
Within the U.S., Washington, D.C. leads in per capita usage at 3.82x its population share, followed closely by Utah at 3.78x—outpacing California, despite its tech dominance. Regional economic characteristics influence usage: California sees high demand for IT support, Florida for financial services, and D.C. for document editing and career assistance.
Countries with higher AI adoption exhibit more diverse applications, spanning education, science, and business operations. In contrast, lower-adoption nations focus heavily on coding—over half of all AI use in India, compared to about one-third globally. Additionally, advanced markets tend to use AI collaboratively (augmentation), while emerging ones lean toward full task delegation (automation), even after adjusting for task types.
This divergence raises concerns about global inequality. If productivity gains accrue primarily to high-adoption regions, AI could widen economic gaps rather than close them, reversing recent trends of convergence.
Enterprise adoption through Anthropic’s first-party API reveals distinct patterns. Unlike consumer-facing use on Claude.ai, API integrations are primarily automated, with 77% of business interactions involving full task delegation, compared to 50% among general users. This reflects the programmatic nature of API access, where AI is embedded directly into software systems.
Coding and administrative tasks dominate enterprise use, while educational and creative applications are less common. Interestingly, frequently used API tasks tend to be more expensive, indicating that businesses prioritize capability and value over cost. This weak price sensitivity suggests that economic impact and feasibility of automation outweigh pricing considerations in deployment decisions.
However, effective deployment in complex domains depends on access to contextual data. Many firms may face bottlenecks due to fragmented or non-digitized information, requiring investments in data infrastructure and organizational restructuring to fully leverage AI.
Anthropic has released the underlying dataset to support independent research on AI’s economic effects. The data includes task-level usage patterns mapped to occupational taxonomies, collaboration modes, and methodological documentation. Currently, geographic breakdowns are available only for Claude.ai traffic.
Key research questions this data can help address include: How does AI adoption affect local labor markets? What drives differences in adoption across regions? Can policy interventions ensure broader distribution of AI benefits? And what determines which tasks can be automated successfully?
Looking ahead, the data suggests AI is enabling new forms of work, not just accelerating existing ones. As models improve and user trust grows, task delegation is increasing. Future reports will further disentangle whether this trend stems from technological advancement or user learning.
— news from Anthropic
— News Original —
Anthropic Economic Index report: Uneven geographic and enterprise AI adoption
AI differs from prior technologies in its unprecedented adoption speed. In the US alone, 40% of employees report using AI at work, up from 20% in 2023 two years ago.1 Such rapid adoption reflects how useful this technology already is for a wide range of applications, its deployability on existing digital infrastructure, and its ease of use—by just typing or speaking—without specialized training. Rapid improvement of frontier AI likely reinforces fast adoption along each of these dimensions.
Historically, new technologies took decades to reach widespread adoption. Electricity took over 30 years to reach farm households after urban electrification. The first mass-market personal computer reached early adopters in 1981, but did not reach the majority of homes in the US for another 20 years. Even the rapidly-adopted internet took around five years to hit adoption rates that AI reached in just two years.2
Why is this? In short, it takes time for new technologies—even transformative ones—to diffuse throughout the economy, for consumer adoption to become less geographically concentrated, and for firms to restructure business operations to best unlock new technical capabilities. Firm adoption, first for a narrow set of tasks, then for more general purpose applications, is an important way that consequential technologies spread and have transformative economic effects.3
In other words, a hallmark of early technological adoption is that it is concentrated—in both a small number of geographic regions and a small number of tasks in firms. As we document in this report, AI adoption appears to be following a similar pattern in the 21st century, albeit on shorter timelines and with greater intensity than the diffusion of technologies in the 20th century.
To study such patterns of early AI adoption, we extend the Anthropic Economic Index along two important dimensions, introducing a geographic analysis of Claude.ai conversations and a first-of-its-kind examination of enterprise API use. We show how Claude usage has evolved over time, how adoption patterns differ across regions, and—for the first time—how firms are deploying frontier AI to solve business problems.
Changing patterns of usage on Claude.ai over time
In the first chapter of this report, we identify notable changes in usage on Claude.ai over the previous eight months, occurring alongside improvements in underlying model capabilities, new product features, and a broadening of the Claude consumer base.
We find:
Education and science usage shares are on the rise: While the use of Claude for coding continues to dominate our total sample at 36%, educational tasks surged from 9.3% to 12.4%, and scientific tasks from 6.3% to 7.2%.
Users are entrusting Claude with more autonomy: “Directive” conversations, where users delegate complete tasks to Claude, jumped from 27% to 39%. We see increased program creation in coding (+4.5pp) and a reduction in debugging (-2.9pp)—suggesting that users might be able to achieve more of their goals in a single exchange.
The geography of AI adoption
For the first time, we release geographic cuts of Claude.ai usage data across 150+ countries and all U.S. states. To study diffusion patterns, we introduce the Anthropic AI Usage Index (AUI) to measure whether Claude.ai use is over- or underrepresented in an economy relative to its working age population.
We find:
The AUI strongly correlates with income across countries: As with previous technologies, we see that AI usage is geographically concentrated. Singapore and Canada are among the highest countries in terms of usage per capita at 4.6x and 2.9x what would be expected based on their population, respectively. In contrast, emerging economies, including Indonesia at 0.36x, India at 0.27x and Nigeria at 0.2x, use Claude less.
In the U.S., local economy factors shape patterns of use: DC leads per-capita usage (3.82x population share), but Utah is close behind (3.78x). We see evidence that regional usage patterns reflect distinctive features of the local economy: For example, elevated use for IT in California, for financial services in Florida, and for document editing and career assistance in DC.
Leading countries have more diverse usage: Lower-adoption countries tend to see more coding usage, while high-adoption regions show diverse applications across education, science, and business. For example, coding tasks are over half of all usage in India versus roughly a third of all usage globally.
High-adoption countries show less automated, more augmented use: 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 uneven geography of early AI adoption raises important questions about economic convergence. Transformative technologies of the late 19th century and the early 20th centuries—widespread electrification, the internal combustion engine, indoor plumbing—not only ushered in the era of modern economic growth but accompanied a large divergence in living standards around the world.4
If the productivity gains are larger for high-adoption economies, current usage patterns suggest that the benefits of AI may concentrate in already-rich regions—possibly increasing global economic inequality and reversing growth convergence seen in recent decades.5
Systematic enterprise deployment of AI
In the final chapter, we present first-of-its-kind insight on a large fraction of our first-party (1P) API traffic, revealing the tasks companies and developers are using Claude to accomplish. Importantly, API users access Claude programmatically, rather than through a web user interface (as with Claude.ai). This shows how early-adopting businesses are deploying frontier AI capabilities.
We find:
1P API usage, while similar to Claude.ai use, differs in specialized ways: Both 1P API usage and Claude.ai usage focus heavily on coding tasks. However, 1P API usage is higher for coding and office/admin tasks, while Claude.ai usage is higher for educational and writing tasks.
1P API usage is automation dominant: 77% of business uses involve automation usage patterns, compared to about 50% for Claude.ai users. This reflects the programmatic nature of API usage.
Capabilities seem to matter more than cost in shaping business deployment: The most-used tasks in our API data tend to cost more than the less frequent ones. Overall, we find evidence of weak price sensitivity. Model capabilities and the economic value of feasibly automating a given task appears to play a larger role in shaping businesses’ usage patterns.
Context constrains sophisticated use: Our analysis suggests that curating the right context for models will be important for high-impact deployments of AI in complex domains. This implies that for some firms costly data modernization and organizational investments to elicit contextual information may be a bottleneck for AI adoption.
Open source data to catalyze independent research
As with previous reports, we have open-sourced the underlying data to support independent research on the economic effects of AI. This comprehensive dataset includes task-level usage patterns for both Claude.ai and 1P API traffic (mapped to the O*NET taxonomy as well as bottom-up categories), collaboration mode breakdowns by task, and detailed documentation of our methodology. At present, geographic usage patterns are only available for Claude.ai traffic.
Key questions we hope this data will help others to investigate include:
What are the local labor market consequences for workers and firms of AI usage & adoption?
What determines AI adoption across countries and within the US? What can be done to ensure that the benefits of AI do not only accrue to already-rich economies?
What role, if any, does cost-per-task play in shaping enterprise deployment patterns?
Why are firms able to automate some tasks and not others? What implications does this have for which types of workers will experience better or worse employment prospects?
Chapter 1: Claude.ai usage over time
Overview
Understanding how AI adoption evolves over time can help predict its economic impacts—from productivity gains to workforce changes. With data spanning from December 2024 and January 2025 (from our first report, ‘V1’) to February and March 2025 (‘V2’) to our newest insights from August 2025 (‘V3’), we can track how AI usage has shifted over the past eight months as capabilities and product features have improved, new kinds of users have adopted the technology, and uses have become more sophisticated. We view the evidence presented below as suggesting that new product features have enabled new forms of work rather than simply accelerating adoption for existing tasks.
How Claude.ai usage for economic tasks has changed
Educational and scientific tasks continue their rise in relative importance
While computer and mathematical tasks still dominate overall usage at 36%, we are seeing sustained growth in knowledge-intensive fields. Educational Instruction and Library tasks rose from 9% in V1 to 12% in V3. Life, Physical, and Social Science tasks increased from 6% to 7%. Meanwhile, the relative share of Business and Financial Operations tasks fell from 6% to 3%, and Management dropped from 5% to 3%.
This divergence suggests AI usage may be diffusing especially quickly among tasks involving knowledge synthesis and explanation, compared to traditional business operations—possibly because these tasks benefit more from Claude’s reasoning capabilities.
New capabilities are shaping usage patterns
At a more granular level, we document changes in task composition that appear linked to features launched between V2 and V3. For example, searching electronic sources and databases grew substantially (0.03% → 0.49%), likely reflecting our web search release in March. In addition, we also see a rise in internet-based research tasks (0.003% → 0.27%), which aligns with the Research mode we released in April.1
We also see other kinds of changes. Tasks relating to developing instructional materials increased by 1.3pp, growing from a base of 0.2% to 1.5%—a more than 6-fold increase that may reflect growing adoption among educators.2 Creating multimedia documents rose 0.4pp, nearly tripling from 0.16% to 0.55%, potentially driven by continued use of our Artifacts feature for building traditional and AI-powered apps within Claude.ai.
Interestingly, the share of tasks involving creating new code more than doubled, increasing by 4.5 percentage points (from 4.1% to 8.6%), while debugging and error correction tasks fell by 2.8 percentage points (from 16.1% to 13.3%)—a net 7.4pp shift toward creation over fixing code. This may suggest that models have become increasingly reliable, such that users spend less time fixing problems and more time creating things in a single interaction.3
Directive automation is accelerating
As in previous reports, we also track not just what people use Claude for but how they collaborate with or delegate to Claude on Claude.ai.
At a high level, we distinguish between automation and augmentation modes of using Claude:
Automation encompasses interaction patterns focused on task completion:
Directive: Users give Claude a task and it completes it with minimal back-and-forth
Feedback Loops: Users automate tasks and provide feedback to Claude as needed
Augmentation focuses on collaborative interaction patterns:
Learning: Users ask Claude for information or explanations about various topics
Task Iteration: Users iterate on tasks collaboratively with Claude
Validation: Users ask Claude for feedback on their work
The share of directive conversations sampled from Claude.ai conversations jumped from 27% in V1 in late 2024 to 39% in V3. This increase came primarily at the expense of task iteration and learning interactions, implying a sizable net increase in the share of conversations exhibiting automative patterns of use – a notable increase in just eight months. This is the first report where automation usage exceeds augmentation usage.
One interpretation is that this is a result of increasing model capabilities. As models improve at anticipating user needs and producing high-quality outputs on first attempts, users may need fewer follow-up refinements. The jump in directive usage could also signal growing confidence in delegating complete tasks to AI, a form of learning-by-doing.4
Whether the growth in directive usage is attributable to improving model capabilities or learning-by-doing could signal very different labor market implications. If more advanced models simply expand the set of automated tasks, then the risk increases that workers performing such tasks will be displaced. However, if instead the rise in directive use reflects learning-by-doing, then workers most able to adapt to new AI-powered workflows are likely to see greater demand and higher wages. In other words, AI may benefit some workers more than others: it may lead to higher wages for those with the greatest ability to adapt to technological change, even as those with lower ability to adapt face job disruption.5 This will be an important area of inquiry for future research.
Looking ahead
The V3 data reveals that AI capabilities and adoption are continuing to progress. Knowledge-based tasks, including educational and scientific applications, continue their fast growth rate, and new product features appear to be enabling different types of work rather than just accelerating existing tasks.
Most strikingly, the data point toward increased delegation of tasks to AI systems–perhaps due to some combination of user trust in the technology as well as improvement of underlying model capabilities. This could also be due to changes in the underlying user base. The next chapter of this report for the first time breaks down usage across geography, allowing us to disentangle temporal vs. geographic changes more clearly going forward. We will continue to track these trends closely in future reports.
Chapter 2: Claude usage across the United States and the globe
Overview
Where AI gets adopted first—and how it’s used—will shape economic outcomes across the world. By analyzing Claude usage patterns across 150+ countries and all US states, we uncover three key dynamics: where early adopters are, what they’re using AI for, and how usage evolves as adoption matures. These geographic patterns provide real-world evidence about AI’s economic diffusion, helping track whether different regions are converging or diverging in their AI adoption, and revealing how local economic characteristics shape technology deployment.
Our data, relying on a privacy-preserving1 analysis of 1 million Claude.ai conversations2, confirmed some of our expectations while challenging others. The US dominates total usage at 21.6%, which is unsurprising given its size and high income. But even when adjusting for the working-age population size, higher-income countries tend to have higher usage. For example, Singapore’s usage rate is 4.5 times what its working-age population would suggest, while large regions of the globe show minimal usage. Interestingly, within the US, DC and Utah outpace California in usage per capita.
We also observe changes in AI use cases as adoption per capita deepens. Countries with lower AI adoption per capita concentrate overwhelmingly on coding tasks—over half of all usage in India, compared to roughly a third globally. As adoption matures, usage diversifies, with a rising emphasis on education, science, and business operations.
Even more striking: mature markets tend to use AI more collaboratively, while emerging markets are more likely to delegate complete tasks to it—perhaps reflecting differences in how AI is deployed by economies at different stages of structural transformation. Our data provides a window into these patterns across geographies, and going forward, will enable us to track whether these adoption gaps narrow, widen, or change in structure over time.
Claude diffusion across the globe
Total Claude usage is highest in the US
Claude adoption overall is highly geographically concentrated. In terms of total global usage, the United States accounts for the highest share (21.6%), with the next highest usage countries showing significantly lower shares (India at 7.2%, Brazil at 3.7%, see Figure 2.1). However, this concentration is affected by the population size of each country3 – larger countries may have larger usage shares purely because of their population size.
Per capita usage of Claude is concentrated in technologically advanced countries
To account for differences in population size, we analyze usage adjusted for the working-age population, introducing a new measure called the Anthropic AI Usage Index (AUI): For each geography, we calculate its share of Claude usage, and its share of the working-age population (ages 15-64). We then calculate the AUI by dividing these shares:
This index reveals whether countries use Claude more or less than expected relative to their working-age population. A region with an AUI > 1 has higher usage than expected after adjusting for population, while a region with an AUI < 1 has lower usage. The results reveal a striking pattern of concentration among small, technologically advanced economies. Israel leads global per capita Claude usage with an Anthropic AI Usage Index of 7 — meaning its working-age population uses Claude 7x more than expected based on its population. Singapore follows at 4.57, while Australia (4.10), New Zealand (4.05) and South Korea (3.73) round out the top five countries in terms of per capita Claude usage. Next, we create per capita usage tiers based on the AUI. We look at countries with at least 200 conversations in our random sample of 1 million conversations, and set thresholds for different usage tiers-based quartiles, i.e. Leading (top 25%), Upper Middle (50-75%), Lower Middle (25%-75%) and Emerging (bottom 25%). We then assign countries, even if they have fewer than 200 observations, to a tier based on their AUI. We assign countries for which we have population data, but no usage in our sample, to a Minimal tier.4 Figure 2.3 illustrates the Anthropic AI Usage Index tiers across the globe, Table 2.1 shows an overview of the tiers and country examples. Zooming into leading and emerging countries in terms of per capita usage This concentration in advanced economies with limited population sizes reflects their established patterns as technology pioneers. For example, both Israel and Singapore rank highly in the Global Innovation Index—a measure of how innovative different economies across the globe are—suggesting that general investments in information technology position economies well for rapid adoption of frontier AI. Overall, these economies can leverage their educated workforces, robust digital infrastructure, and innovation-friendly policies to create fertile conditions for AI. Notable is the position of major developed economies in Claude usage. The United States (3.62) ranks among leading countries in terms of per capita adoption, with Canada (2.91) and the United Kingdom (2.67) having elevated but more moderate rates of adoption as compared to their population. Other major economies show lower adoption, including France at 1.94, Japan at 1.86, and Germany at 1.84. Meanwhile, many lower and middle-income economies show minimal Claude usage, with many countries across Africa, Latin America, and parts of Asia showing Claude adoption below what would be expected based on their working-age population. This includes Bolivia (0.48), Indonesia (0.36), India (0.27), and Nigeria (0.2). This variation in usage is reflective of income differences across these economies. We see a strong positive correlation between Claude adoption and Gross Domestic Product per working-age capita (see Figure 2.4), with a 1% increase in GDP per capita being associated with a 0.7% increase in Claude usage per capita. The disparities in Claude usage likely reflect a confluence of factors, some of which are correlated with income: Digital infrastructure: High-usage countries typically have robust internet connectivity and cloud computing access needed to access AI assistants. Economic structure: As documented in this and previous reports, Claude capabilities are well-suited to various tasks typical of knowledge workers. Advanced economies tend to have a greater share of the workforce in such roles as compared to lower-income economies with a larger employment share in manufacturing. Regulatory environment: Governments differ in how actively they encourage the use of AI across different industries and in how heavily they regulate the technology. Awareness and access: Countries with stronger connections to Silicon Valley and AI research communities may have greater awareness of and access to Claude. Trust and comfort: Public opinion on trust in AI varies substantially across countries. Claude diffusion across the United States Within the US, California overwhelmingly leads with 25.3% of usage. Other states with major tech centers like New York (9.3%), Texas (6.7%), and Virginia (4.0%) also rank highly. Though not adjusted for population, we suspect that these strong adoption figures partly reflect rapid adoption in technology hubs—in keeping with how economically consequential technologies have historically tended to diffuse. This narrative becomes more complex, however, when we adjust for the population size of each state. Surprisingly, the District of Columbia leads with an Anthropic AI Usage Index of 3.82, indicating that Claude usage in DC is 3.82x greater than its share of the country’s working-age population. Closely following is Utah (3.78), notably ahead of California (2.13), New York (1.58) and Virginia (1.57).7 We document a similar, but weaker correlation than at the global level between Claude adoption and income per capita across US states. Income differences explain less than half the variation in cross-state adoption rates. Despite this weaker correlation, we find that Claude adoption rises faster with income: Each 1% increase in state GDP per capita is associated with a 1.8% increase in the AI Usage Index. Task usage patterns across countries We observe notable variation in how Claude is used in different countries. As in past reports, we analyze these trends using two different approaches. First, we classify conversations into tasks according to O*NET, a US taxonomy that maps specific tasks to occupations and occupation groups (e.g., a task involving software debugging would fall into the Computer and Mathematical occupation group). Second, we use Claude to construct a bottom-up taxonomy of user requests on Claude.ai, which provides insight into usage patterns that do not fit neatly into existing taxonomies. For example, the request cluster “help write and improve cover letters for job applications” (lowest level) feeds into the higher-level cluster “help with job applications, resumes, and career documents” (middle level), which in turn feeds into the cluster “help with job applications, resumes, and career advancement” (highest level). These two complementary approaches allow us to both report results aligned with standard labor statistics, and provide flexibility to capture tasks that standard taxonomies miss. Higher per capita Claude usage is associated with more diverse task usage When analyzing O*NET tasks aggregated at the highest level (in terms of the Standard Occupation Classification occupation groups they belong to), we notice strong variation across countries. While the overall pattern is noisy–especially for countries with fewer observations–Figure 2.7 suggests that as we progress from lower to higher per capita Claude adoption, usage shifts away from tasks in the Computer and Mathematical occupation group (e.g., programming) to more diverse tasks in areas such as education, office and administrative uses, and arts. We also see increased usage in the life, physical and social sciences. Country idiosyncrasies also emerge when looking at our bottom-up request taxonomy.8 Take, for example, the United States, Brazil, Vietnam, and India, which represent the country with the highest total usage within a given Anthropic AI Usage Index tier. Users in the United States disproportionately use Claude for household management purposes, to search for jobs, and for medical guidance compared to the global average. By contrast, Claude users in Brazil have comparatively high usage for both translation and legal services. Vietnam’s top disproportionate requests are related to software development and education, and India’s top disproportionate requests focus almost exclusively on software development. This likely reflects local specialization: Brazil has been an early adopter of AI in the judicial system, and India has a large information technology sector. Across all countries, software development emerges as the most common use of Claude. Why do developer tasks consistently lead in overall Claude usage patterns? Several factors likely contribute to this effect: Model-task fit: Claude is a very strong coding model and readily deployed across code generation, debugging, and technical problem-solving tasks. Developer receptivity: Developer communities embrace new tools rapidly, and this usage diffuses through their social and professional networks. Low organizational barriers: Individual developers can typically adopt Claude without complex approval processes—in contrast to, say, medical use cases. Task usage patterns across the United States In this section we explore patterns of Claude usage across states within the US, giving us further insight into how local economic conditions shape usage patterns. As we discuss above, cross-state differences in the Anthropic AI Usage Index account for less than half of the variation in income differences across US states. This suggests that other regional differences—including the compatibility of Claude capabilities with the occupational composition of the local workforce—play a larger role in determining why usage is more concentrated in some states than others. In a number of states, we see evidence that local patterns of AI use aligns with distinctive features of the local economy. When analyzing the top states in each usage tier—California for leading, Texas for upper middle, Florida for lower middle, and South Carolina for emerging—we see strong variation in terms of our bottom-up request taxonomy (see Figure 2.9). For example, California shows disproportionate use for IT-related requests, digital marketing and translation, likely reflecting its tech sector and linguistically diverse population. California also has disproportionately frequent requests for help with basic numerical tasks, which may represent tests of model capabilities or abuse. Florida sees disproportionate use for business advice and fitness, potentially tied to its role as a financial hub with relatively low tax rates and a warm climate amenable to outdoor activities. Within the US, D.C. leads in terms of per capita Claude usage, with a disproportionate focus on document editing, information provision and job applications across both the O*NET task classification and bottom-up categorization (see Figure 2.10). For example, help with job applications is 1.84x as common in DC as in the US overall. Our interactive dashboard allows everyone to explore the full range of variation and patterns across US states. Geographic patterns in human-AI collaboration While previous sections examined what tasks people use Claude for, an equally revealing pattern emerges in how they interact with it. Here, we use the same augmentation and automation collaboration patterns as defined in Chapter 1. Countries have different task mixes, meaning that they focus on different economic tasks, which may partly explain differences in automation patterns. In this section, we investigate whether automated use is systematically different among low and high per capita adoption economies—even when controlling for differences in task mix.9 We find that even when controlling for the task mix of a country, users from different countries show notably different preferences for autonomous delegation versus collaborative interaction. As Claude usage per capita increases, countries shift from automation-focused to augmentation-focused usage. This is somewhat counter-intuitive, since we are controlling for the more diverse task composition across different countries. We speculate that cultural and economic factors might affect the automation share, or perhaps that early adopters in each country tend to use AI in a more automotive way—but more research is needed here. Conclusion Our analysis of Claude usage patterns across geographies reveals several key insights. One of the most striking is the geographic concentration of Claude usage. The leadership of the US and California in terms of Claude usage overall, and the strong correlation of Claude usage and income per capita, suggest parallels to past technologies in which initial geographic concentration and specialized use were a key feature. Drawing parallels to the diffusion patterns of prior technologies may help us better understand the diffusion and impact of AI. Surprisingly, geography shapes not just what AI tools are used for, but how they are used. Users in economies with relatively low per capita usage have a relative preference for delegating tasks to Claude (automation), whereas users in economies with high per capita usage are somewhat more likely to prefer more collaborative or learning-based interactions with Claude (augmentation), even when controlling for the task mix. Similar to the local specialization in task use, the local specialization in AI collaboration patterns suggests that impact of AI could be very different in different regions. The geographic patterns of AI adoption—where it is used, for which tasks, and how—suggest that in order to realize the potential of AI to benefit people across the globe, policymakers need to pay attention to local concentration of AI use and adoption, and address the risk of deepening digital divides.