Essential Points
- Users on Claude for 6+ months show 3 to 5 percentage points higher conversation success rates than newer users
- Average task value on Claude.ai fell from $49.30 to $47.90 as casual, lower-complexity use expanded rapidly
- Business sales automation and automated trading workflows at least doubled in API share since November 2025
- Top 20 countries account for 48% of all per-capita Claude usage, up from 45%, as the global adoption gap widens
About 49% of jobs have had at least a quarter of their tasks performed using Claude, yet the workers gaining the most from AI are those who started earliest. Anthropic’s March 2026 Economic Index, published March 24, 2026 and covering Claude usage from February 5 to 12, documents a clear pattern: experience with AI compounds into measurable advantage, and that gap is widening. The report’s central finding, called the “learning curve,” has direct consequences for how AI reshapes the labor market and who benefits most.
Use Cases Diversified, But Average Task Value Dropped
Between November 2025 and February 2026, Claude.ai usage spread across a broader range of tasks. The top 10 most common O*NET tasks fell from 24% of all conversations to 19%, partly because coding tasks migrated from Claude.ai to the first-party API, where Claude Code splits work into many smaller, distinct task calls.
The mix shift also reflects changing user demographics. Personal use rose from 35% to 42% of conversations, while coursework fell from 19% to 12%, a drop partially explained by winter academic breaks in key markets. At the same time, increasing signups around February brought more casual users.
The practical result: the average hourly wage associated with Claude.ai tasks fell from $49.30 to $47.90, driven by a rise in simple factual queries such as sports outcomes and weather checks. The average years of education required for human inputs declined from 12.2 to 11.9 years. The platform is reaching more people, but those people are using it for progressively simpler tasks.
API Usage Tells the Opposite Story
On the first-party API, task complexity moved in the other direction. Since August 2025, the share of Computer and Mathematical tasks in the API increased by 14%, while decreasing by 18% in Claude.ai. The top 10 O*NET tasks in the API now account for 33% of traffic, up from 28% since August 2025.
Anthropic frames this migration as a leading indicator of labor market impact. When tasks move from a consumer interface to automated API workflows, those workflows are far more likely to be directive, with less need for a human in the loop. The associated jobs face more imminent transformation as a result.
Augmentation, the collaborative mode where AI complements rather than replaces human decisions, increased slightly on Claude.ai, driven by small increases in validation and learning patterns. Automation decreased sharply in the API data, though Anthropic notes this reflects changes in task composition rather than a structural retreat from automation.
Two Workflows at Least Doubled in API Share
Two specific API workflow categories showed at least 2x growth between the November 2025 and February 2026 samples.
The first is business sales and outreach automation, covering sales enablement generation, B2B lead qualification research, customer data enrichment, and cold-email drafting. The second is automated trading and market operations, including monitoring markets or positions, proposing specific investments, and informing traders of market conditions.
Both categories involve directive, low-human-oversight workflows. Anthropic’s previous labor market impact research identified Customer Service Representatives as having high exposure to this automation pattern, with Claude performing a high share of their tasks in automated workflows.
Experienced Users Achieve Meaningfully Better Results
The most important finding in the March 2026 report is not about usage volume. It is about outcomes.
Users who have been on Claude for six months or more are approximately 5 percentage points more likely to have a successful conversation in a simple bivariate comparison. After controlling for specific O*NET tasks, request clusters, model choice, use case, and country, high-tenure users still show a 3 to 4 percentage point higher success rate. Anthropic notes this is not explained by higher-tenure users simply bringing different kinds of tasks.
High-tenure users also differ in how they work. They are 7 percentage points more likely to use Claude for work, bring higher-education tasks to Claude, and use Claude across a wider range of applications. The top 10 O*NET tasks account for 20.7% of usage for high-tenure users versus 22.2% for newer users, reflecting broader and more varied application. High-tenure users are also more likely to iterate with Claude collaboratively rather than delegate tasks entirely.
The tasks with the highest average user tenure include AI research, git operations, revising manuscripts, and startup fundraising. The tasks with the lowest average tenure include writing haikus, checking sports scores, and suggesting food for a party.
Users Match Model Power to Task Complexity
Claude users are calibrating model selection to task value, not selecting randomly. Among paid Claude.ai accounts, which have access to all model classes, 55% of Computer and Mathematical tasks use Opus compared to 45% of Educational tasks. Opus is used 4 percentage points more than average for coding tasks and 7 percentage points less than average for tutoring-related tasks.
For every additional $10 of hourly wage associated with a task, Opus usage increases by 1.5 percentage points on Claude.ai and 2.8 percentage points via the API. API users show approximately twice the sensitivity to task complexity compared to Claude.ai users, which Anthropic attributes to programmatic workflow operators having greater reason to switch between models.
For tech professionals and knowledge workers in the US and India, this data carries a direct implication: learning which model to deploy for which task affects both the cost and quality of AI-assisted work.
US Convergence Continues, Global Gap Widens
Within the United States, AI usage per capita continued to converge across states. The share of per-person usage going to the top five states dropped from 30% to 24% between August 2025 and February 2026. However, the pace of convergence slowed. Updated estimates now project that US states could reach roughly equal per-capita usage in 5 to 9 years, compared to the 2 to 5 year estimate from the previous report.
Globally, the pattern reversed. The Gini coefficient for country-level usage rose over the same period. The top 20 countries now account for 48% of all per-capita Claude usage, up from 45%. Countries using Claude the most per capita account for a larger share of overall usage, a trend that Anthropic describes as a persistent gap in global adoption.
What the Learning Curve Means for the Labor Market
Anthropic’s report identifies a specific mechanism through which AI may deepen existing inequality: early adopters with high-skill tasks have more successful interactions with Claude than later, less technical adopters. These early adopters may simultaneously be the most exposed to AI-driven disruption and the most aided by AI in this initial, augmentative wave of adoption.
Economists describe this as skill-biased technological change, where innovation raises wages for high-skill workers while depressing them for others. The March 2026 data offers the clearest evidence yet that this channel is already operating in AI adoption, not as a future risk but as a present, measurable dynamic.
Limitations and Honest Considerations
The report covers one week of data (February 5 to 12) and coincided with the release of Claude Opus 4.6 and Anthropic’s Super Bowl advertising campaign, which brought a surge of first-time casual users. This timing inflates the apparent decline in average task complexity. The tenure-based findings also face acknowledged survivorship bias: users still active after six months may simply be those whose tasks AI handles well, not necessarily those who have learned more. Anthropic notes this limitation directly and commits to cleaner cohort isolation in future reports.
Claude Just Turned Your Phone Into an AI Command Center While Millions Walked Away From ChatGPT
Frequently Asked Questions (FAQs)
What is the Anthropic Economic Index?
The Anthropic Economic Index is a recurring research report that uses privacy-preserving data analysis to track how Claude is used across the economy. It studies task types, interaction modes, and labor market implications. The March 2026 edition covers a sample of 1 million conversations from February 5 to 12, 2026.
Do experienced Claude users actually perform better?
Yes. Users on Claude for six months or more show approximately 5 percentage points higher conversation success rates in bivariate comparisons, and 3 to 4 percentage points higher after controlling for task type, country, model selection, and use case. Anthropic identifies this as consistent with learning-by-doing rather than simple task selection effects.
What types of work are growing fastest on Claude’s API?
Two API workflow categories at least doubled in usage share between November 2025 and February 2026: business sales and outreach automation (B2B lead research, cold emails, data enrichment) and automated trading and market operations (position monitoring, investment proposals, trader alerts). Both involve directive, low-human-oversight workflows.
Is AI use growing in India and developing countries?
Globally, per-capita usage has become more concentrated. The top 20 countries now account for 48% of usage, up from 45%, and the Gini coefficient for country-level usage rose between August 2025 and February 2026. Within the US, convergence across states continued, though at a slower pace than the previous report.
Why did the average task value on Claude.ai drop?
The average hourly wage associated with Claude.ai tasks fell from $49.30 to $47.90 between November 2025 and February 2026. The decline reflects broader user adoption bringing simpler queries: sports outcomes, weather checks, and product comparisons. High-complexity coding work has largely migrated to the API, where average task value is rising.
Does selecting a more powerful Claude model improve outcomes?
Users who select Opus for high-value, complex tasks show alignment between model capability and task difficulty. For every $10 increase in the hourly wage associated with a task, Opus usage rises 1.5 percentage points on Claude.ai and 2.8 percentage points via the API. Calibrating model choice to task complexity is itself a learned skill that distinguishes experienced users.
What does the report mean for workers concerned about job displacement?
The data shows AI is currently augmenting most high-skill workers rather than replacing them, but automation is accelerating in customer service and sales workflows. The report explicitly identifies a channel through which skill-biased technological change may already be unfolding: early adopters benefit most from AI while simultaneously facing the highest exposure to eventual disruption.
