Anthropic: Claude's Evolution Could Widen Inequalities

An In-Depth Analysis of the Economic Impact of Claude

Anthropic has recently published its fifth report on the Economic Index, based on data collected in February 2026. This report highlights how the use of Claude, an artificial intelligence model, is evolving within the economy. One of the major findings is that the more users interact with Claude, the better their outcomes, which could potentially widen existing economic inequalities.

The report employs a privacy-respecting analysis system to examine the use of Claude without revealing the content of individual conversations. The sample includes one million conversations from Claude.ai and the Anthropic API.

Diversification and Simplification in the Use of Claude

Since the first report in November 2025, the use of Claude.ai has seen significant diversification. In February, the ten most common tasks accounted for only 19% of total traffic, down from 24% three months prior. Programming remains the primary application, making up 35% of usage, but it is increasingly shifting to the API, where Claude Code is taking a growing share.

The share of personal requests has also increased, rising from 35% to 42%. However, the average economic value of tasks performed on Claude.ai, measured by the hourly wage of American workers in related professions, has slightly decreased from $49.30 to $47.90.

According to Anthropic, this trend fits within a typical adoption curve, where early users prefer specialized tasks like programming, while later users bring a broader range of tasks, including sports scores, product comparisons, and home interview questions. Overall, about 49% of all professions have at least a quarter of their tasks performed via Claude.

Increased Collaboration Between Experienced Users and Claude

The report distinguishes between automation, where Claude operates primarily autonomously, and augmentation, where humans collaborate with the model. Augmentation has slightly increased on Claude.ai.

The gap between experienced users and newcomers is notable. Experienced users are 8.7 percentage points less likely to simply give an instruction to Claude and are much more inclined to iterate on tasks. They use Claude 7 percentage points more often for professional purposes and bring more complex requests.

At the top of the experience scale, activities include AI research, Git operations, and manuscript review. In contrast, newcomers tend to request haikus, sports scores, or food suggestions for parties.

Progression of AI Skills with Practice

Even after statistically controlling for task type, model choice, use case, and country of origin, the effect of experience persists. Experienced users see a success rate about 4 percentage points higher than newcomers working on the same task. In other words, getting good results from AI platforms is a skill that improves with practice.

Anthropic measures success by asking Claude to evaluate anonymized transcripts to determine if a conversation has achieved its goal. The experience effect amounts to about four percentage points.

User Preferences for the Opus Model

For the first time, the report also examines which models people choose. Paying users of Claude.ai are gravitating towards Opus, the top-performing option, specifically for complex work. For programming tasks, 55% choose Opus; for educational tasks, only 45% do. API users respond even more strongly to task complexity in their model choice—about twice as much—which makes sense given that the API audience is more technical than the average web user of Claude.

Growth of Automation in Sales and Financial Trading

In the API, the report notes two categories of workflows that have at least doubled their share since November. The first is sales automation and client prospecting: B2B lead qualification, email generation for cold calls, etc. The second concerns automated trading operations, including market monitoring and specific investment recommendations.

In the United States, usage is still converging between states, but more slowly than before. Anthropic now estimates that it will take 5 to 9 years for per-person usage to stabilize between states, compared to a previous projection of 2 to 5 years. Internationally, the gap is actually widening. The 20 countries with the highest per-person usage now account for 48% of traffic adjusted for population, up from 45% in the previous report.

Early Users and the Widening Inequalities

The report refers to the economic concept of "skill-biased technological change." Early users working on technically demanding tasks derive more from their interactions with Claude and benefit the most, while also being the group most exposed to AI-induced disruption.

The authors conclude with a warning about the labor market: if using AI effectively is a skill that develops over time, the advantages of early adoption could become self-reinforcing. The data from the report is available on Hugging Face.