Does longer use of AI by employees lead to higher work skills and efficiency? Claude’s application scope has expanded from early tech adopters to mainstream users, and the nature of AI-assisted work is changing. Its impact on the labor market and economic inequality is also evolving. This article is compiled from the latest Anthropic Economic Index (Human Economy Index Report), which provides a research report on AI integration into the modern workforce. The report is based on data from February 5 to 12, 2026, drawn from one million conversation samples between Claude.ai consumers and API developers, tracking usage changes and geographic adoption trends. It offers an analysis of user AI usage trends and future economic return forecasts.
Programming remains the primary use, with increased daily personal interactions
The most noticeable signal in the data is that Claude’s user base is expanding from a core group of early tech enthusiasts to a broader audience. Programming still dominates, with computer and math-related professions accounting for 35% of Claude.ai conversations, but the concentration on specific tasks is decreasing significantly.
This shift partly results from coding work migrating to APIs, especially through Claude Code, which breaks down programming tasks into multiple smaller API calls. However, this diversification also reflects a genuine expansion of the user base: conversations related to personal use increased from 35% to 42% of Claude.ai traffic, mainly due to sports inquiries, product comparisons, and home maintenance questions. Conversely, conversations related to coursework dropped from 19% to 12%, partly because some countries are on winter break.
Average hourly wages for Claude users have slightly decreased but remain above typical wages
More general users are starting to use Claude, and the average task valuation (measured by the average hourly wage of workers in the US performing these tasks) on the Claude.ai platform has slightly declined from $49.30 to $47.90 per hour. This aligns with the classic technology adoption curve: early users prioritize high-value tasks (like software development), while later users apply it to broader, simpler daily tasks. Despite the slight decline, Claude users are still engaged in tasks requiring education and wages above the US average, highlighting that AI adoption remains concentrated among knowledge workers.
Other indicators also show a slight decrease in Claude.ai’s complexity: the average education years required for user inputs dropped from 12.2 to 11.9 years; users are granting AI greater autonomy, with estimated task completion times by humans decreasing by about 2 minutes.
API automation continues to increase
Although Claude.ai’s development trend is toward more augmented applications (AI-assisted rather than replacing human work), its API development is moving in the opposite direction. Compared to November 2025, by February 2026, API usage in two specific workflow categories has more than doubled.
Sales and outreach automation: generating sales data, qualifying B2B leads, enriching customer data, and composing cold emails.
Market monitoring, investment advice, real-time trading alerts, automated trading, and market operations.
These findings indicate that automation-focused use cases are developing faster in the developer ecosystem than in consumer environments. This pattern has significant implications for industries like sales, finance, and customer service. The report notes that as coding tasks shift from Claude.ai to APIs, these jobs may face more urgent transformations. Increased API automation is seen as an early indicator of occupational change.
US regional gap narrows, global gap widens
Within the US, usage trends are reducing regional disparities: the per capita usage rate in the top five states dropped from 30% in August 2025 to 24% in February 2026. The baseline usage in various states is also declining, indicating that states with lower adoption are catching up. However, the pace has slowed; at current rates, achieving balanced per capita usage across states could take 5 to 9 years, rather than the previously estimated 2 to 5 years.
Globally, the disparity is increasing: on the international level, the trend is less optimistic. The inequality in global AI usage is worsening, with the top 20 countries’ per capita usage share rising from 45% to 48%. The gap between high-income, well-connected countries is widening further, raising concerns about the growing global AI divide.
The uneven global AI application reflects broader “AI economic inequality” patterns. If early adopters in high-income countries disproportionately benefit from productivity gains, existing economic gaps could be exacerbated.
Experience makes you more proficient with AI
One of the most notable and policy-relevant findings in this report is the close link between user experience and AI effectiveness. The research team analyzed usage patterns across different user groups (based on platform usage duration), comparing “experienced users” (registered for at least six months) with new users.
Experienced users work differently
Long-term users not only use Claude more frequently but also more effectively. They tend to collaborate iteratively with Claude, propose more complex tasks, and require fewer revisions to get the information they need.
Main tasks include AI research, Git operations, manuscript editing, and startup funding
The report finds that for every additional year of experience using Claude, the educational complexity of their inputs increases by nearly a year’s worth of learning. This indicates that users are genuinely improving their AI prompt skills over time, not just asking the same questions more often.
After controlling for task type, language, model choice, and user region, experienced users still outperform newcomers by 4 percentage points in success rate during conversations. This strongly suggests that practice and experience with AI translate into more effective applications.
Data also shows that experienced users tend to self-select into tasks that require more engagement. The tasks with the longest average user engagement include AI research, Git operations, manuscript editing, and startup funding. Conversely, tasks with the shortest engagement times include creating haikus, checking sports scores, and food recommendations—typical leisure exploration activities.
API developers use Opus for computer and math projects
The report also indicates that users—especially API developers—are becoming more cautious in deploying models. For Claude.ai users, Opus (the most powerful model category) is chosen in 55% of computer and math tasks but only 45% of educational tasks. As task value increases by $10/hour, the likelihood of Claude.ai users choosing Opus increases by 1.5 percentage points, while API users’ choice rises by 2.8 percentage points.
In developer APIs, the calibration between models and tasks is twice as high as in consumer products, indicating that professionals optimize for both cost and functionality.
AI experience as a career advantage
The findings on the learning curve are perhaps the most impactful: if experienced AI users consistently outperform less experienced ones on the same tasks, AI tools may be widening rather than narrowing the skills gap. Employees who adopt AI early and have used it for months or years may have established lasting productivity advantages.
This aligns with what economists call “skill-biased technological change”: new technologies increase wages for high-skilled workers while potentially displacing low-skilled jobs. Those most vulnerable to AI disruption might also be the ones who benefit most from it.
API automation is quietly accelerating
The doubling of automation workflows in sales and trading is a significant signal. These are not hypothetical use cases but real, large-scale deployments. Previous reports highlighted the reliance of sales and customer service roles on AI; now, these roles are gradually achieving concrete automation workflows. Policymakers and workforce planners should pay close attention to this trend.
AI adoption is progressing toward wealthier nations domestically, but widening globally
While the US is moving toward convergence, the global AI adoption trend is heading in the opposite direction. The share of Claude usage per capita in high-income countries continues to grow. If productivity gains from AI are concentrated in already wealthy economies, the resulting international inequality could be severe, especially for economies unprepared to leverage AI tools at scale.
Businesses should seize the opportunity to start AI literacy training
For companies, the most practical AI deployment strategy may be the simplest: time and practice matter. If experience correlates strongly with success, providing employees with ongoing structured projects, prompt strategies, and use case training can significantly boost productivity. Data shows that AI proficiency is not innate but cultivated through continuous use.
What the Human Economy Index means for employment and the economy
As AI becomes more widespread and early developers’ advantages diminish, early adopters tend to focus on high-value, high-skill applications. As technology reaches more people, AI is increasingly integrated into daily life. This is a healthy sign of mature technology adoption but also a warning: the window for early AI developers to gain a competitive edge may be closing.
Who benefits most from AI? When?
The economic returns from AI are not evenly distributed. Experienced, skilled, and continuously investing individuals reap the greatest rewards. As AI becomes a general tool for knowledge workers, enhancing organizational AI capabilities may become the most important economic policy in the next decade.
Early AI users are not just first adopters—they are becoming the most effective users. In a world where AI can significantly boost productivity for those who know how to use it, “AI experience” is the new competitive advantage. The March 2026 Human Economy Index report paints a complex picture of an AI-enabled economic transformation, with tools like Claude increasingly integrated into the workplace.
Source: Human Economy Index Report—Learning Curve (March 24, 2026)
Original report: anthropic.com/research/economic-index-march-2026-report
Authors: Maxim Massenkoff, Eva Lyubich, Peter McCrory, Ruth Appel, Ryan Heller
