Anthropic’s latest data on Claude points to a simple but important shift: AI use is becoming broader, more personal, and more uneven. The fifth Economic Index Report, based on data from February 2026, suggests that people who have spent more time with Claude are not just using it more often. They are using it differently, and with better results.
The report is based on one million conversations from Claude.ai and Anthropic's first-party API. Anthropic says it used a privacy-compliant analysis system that examines usage patterns without revealing the content of individual conversations.
Claude use is spreading beyond early specialist tasks
Claude.ai activity has become more diversified since the first report in November 2025. The ten most common tasks made up 19 percent of traffic in February, compared with 24 percent three months earlier. That means usage is less concentrated around a narrow set of tasks than it was before.
Coding remains the largest application at 35%, but the report says it is increasingly moving toward the API, where Claude Code makes up a growing share. On Claude.ai itself, the mix is shifting toward a broader range of everyday requests.
Personal requests rose from 35 to 42 percent. The average economic value of tasks completed on Claude.ai, measured by the hourly wage of US workers in related professions, fell slightly from 49.30 to 47.90 dollars.
Anthropic frames this as a typical adoption curve. Early users often focus on specialized work such as programming. Later users bring more varied needs, including sports scores, product comparisons, and home maintenance questions.
The report also says around 49 percent of all professions have at least a quarter of their tasks carried out via Claude. That does not mean every profession is using Claude in the same way. It does show that AI assistance is no longer confined to a few technical categories.
Experienced users work with Claude differently
A major distinction in the report is between automation and augmentation. In automation, Claude largely works on its own after receiving an instruction. In augmentation, the person and the model work together through a more interactive process.
On Claude.ai, augmentation increased slightly. The more striking finding is the gap between new and experienced users. Veterans are 8.7 percentage points less likely to simply give Claude an instruction and more likely to iterate on tasks.
That matters because iteration changes the role of the model. Instead of treating Claude as a tool that receives a single command, experienced users appear to treat it more like a collaborator that can be guided, corrected, and refined over several turns.
The report also finds that experienced users use Claude 7 percentage points more often for professional purposes. They bring more complex requests as well. At the high end of the experience scale, Anthropic identifies activities such as AI research, Git operations, and manuscript revision.
Newcomers tend to use Claude for simpler or more casual requests, including haikus, sports scores, or party food suggestions. The contrast does not only reflect different interests. It also suggests a learning curve in how people discover where the model is useful and how to ask for work that benefits from repeated interaction.
Better AI results appear to come with practice
The report’s most important labor-market implication is that effective AI use looks like a skill. Even after statistically controlling for task type, model choice, use case, and country of origin, experienced users still see a success rate roughly 4 percentage points higher than newcomers working on the same task.
That finding is modest in size but large in meaning. If two people use the same AI system for the same kind of work, the person with more experience may get better outcomes simply because they know how to work with the system more effectively.
The report also looks at model choice for the first time. Paying Claude.ai users tend to choose Opus, the most capable option, for complex work. For coding, 55 percent choose Opus; for educational tasks, only 45 percent do.
API users are even more responsive to task complexity when choosing a model, reacting roughly twice as much. The source notes that this fits with an API audience that is more technical than the average Claude web user.
Anthropic’s authors also acknowledge that cohort effects are likely involved. Early adopters were probably more tech-savvy from the start, and users who continued using Claude after a year may have already found the kinds of tasks where the model performs well.
The API is showing fast growth in specific workflows
The API data highlights two workflow categories whose share has at least doubled since November. One is sales and customer outreach automation, including B2B lead qualification and cold-call email generation.
The other is automated trading operations, including market monitoring and specific investment recommendations. The report does not present these as the only fast-growing API uses, but it flags them as notable areas of expansion.
The geographic picture is also uneven. Within the U.S., usage is still converging across states, but the pace has slowed. Anthropic now estimates it will take 5 to 9 years for usage per person to level out between states, compared with an earlier projection of 2 to 5 years.
Internationally, the gap is widening. The 20 countries with the highest usage per person now account for 48 percent of population-adjusted traffic, up from 45 percent in the previous report.
Early AI adoption may become self-reinforcing
The report connects these findings to the economic concept of "skill-biased technological change." The concern is that people who adopt AI early, especially for technically demanding tasks, may get more value from the technology while also becoming better at using it.
That creates a possible feedback loop. More experienced users receive better results, which can make them more likely to use Claude for professional work, which gives them still more practice. Meanwhile, newer users may remain focused on simpler tasks and see less of the productivity upside.
The same group that benefits most may also be highly exposed to AI-driven disruption. That is the tension at the center of the report: technical users are positioned to gain from Claude, but they may also work in areas where AI changes the shape of labor most directly.
Anthropic’s warning is that, if AI skill builds over time, the benefits of early adoption could become self-reinforcing. Access to the model is only one part of the story. The ability to use it well may become an advantage of its own.
The report’s data is available on Hugging Face. Earlier Economic Index work found that AI assists humans more often than it replaces their work, and that 36 percent of all occupations already use AI for at least a quarter of their tasks.