% of all job categories now use Claude for at least a quarter of tasks, indicating AI became workplace infrastructure, not niche tool.
Top 10 tasks dropped from 24% to 19% of conversations, showing AI use diversifying across more applications rather than deepening in same areas.
Users with 6+ months experience show 10% higher success rates and use Claude collaboratively as thinking partner versus newer users' one-shot directive approach.
Personal use surged from 35% to 42% while coursework dropped, driven by casual queries like sports scores flooding in from growing user base.
API automated business workflows doubled while serious programmatic access users select models deliberately based on task complexity, unlike casual Claude.ai users.
Anthropic just dropped their March 2026 Economic Index report. It's based on 1 million sampled conversations from Claude.ai and the API, collected between February 5-12, 2026. Three months after Opus 4.5 launched, right as Opus 4.6 went live. Here's what the data actually says.
The Big Picture: AI Is Spreading, Not Deepening
The most interesting finding isn't about any single industry. It's about diversification. The top 10 tasks on Claude.ai dropped from 24% of all conversations in November 2025 to 19% in February 2026. People are using AI for more things, not just the same things more often.
Coding still dominates at 35% of Claude.ai conversations. But 49% of all job categories have now seen at least a quarter of their tasks performed using Claude. That's not a niche tool anymore. That's infrastructure.
Who's Using Claude for What
Personal use rose from 35% to 42%. Coursework dropped from 19% to 12%. The average task value actually declined slightly, from $49.30 to $47.90 hourly wage equivalent. Not because people are doing less valuable work, but because simpler queries are flooding in. Sports scores, weather, product comparisons. The casual user base is growing fast.
On the API side, it's a different story. Automated business workflows doubled in frequency. Lead qualification, cold-email drafting, position monitoring, investment proposals. Computer and mathematical tasks increased 14% on the API while decreasing 18% on Claude.ai. The serious automation is moving to programmatic access.
The Experience Gap Is Real
This is the section that should concern anyone not investing time in learning these tools. Users with 6+ months of experience show:
7 percentage points more work-related usage
10% fewer personal (casual) conversations
6% higher education level in their inputs
10% higher conversation success rate (even after controlling for task selection, country, model, and language)
Experienced users tend to use Claude more collaboratively. They iterate. They refine. They treat it as a thinking partner, not a search engine. Newer users are more likely to use directive, one-shot prompts.
The tasks tell the story clearly. High-tenure users: AI research, git operations, manuscript revision, startup fundraising. Low-tenure users: haiku writing, sports scores, meal suggestions.
Model Selection: People Self-Sort
Opus usage correlates directly with task complexity. For every $10 increase in hourly wage equivalent of a task, Opus adoption jumps 1.5 percentage points on Claude.ai and 2.8 points on the API. Software developers use Opus 34% of the time. Tutoring tasks: 12%.
API users are roughly twice as responsive to task complexity when choosing models. They're more deliberate. Makes sense. When you're paying per token in production, you think about which model fits the job.
Geography: The Gap Is Growing Internationally
Within the US, adoption is slowly spreading. The top five states went from 30% of usage in August 2025 to 24% in February 2026. But the convergence is decelerating. What was projected as a 2-5 year equalization is now looking like 5-9 years.
Internationally, it's worse. The top 20 countries now account for 48% of per-capita usage, up from 45%. Global AI adoption is concentrating among nations that already had high adoption. The rich-get-richer dynamic is real.
What This Means for Builders
Three takeaways if you're building with AI:
1. The learning curve matters more than the tool. A 10% success rate gap between experienced and new users is massive. Time invested in understanding how to work with AI compounds. Setting up your environment, learning prompt patterns, building workflows. It's not optional anymore.
2. Automation is moving to the API. The casual Claude.ai interface is becoming more consumer-oriented. The serious work is happening programmatically. If you're still doing everything through the chat interface, you're leaving capability on the table.
3. Diversification is the trend. AI isn't replacing one type of job. It's touching everything, a little bit at a time. 49% of job categories with at least 25% task exposure means every industry needs to pay attention, not just tech.
The full dataset is available on Hugging Face if you want to dig into the numbers yourself.
The gap between people who know how to use these tools and people who don't is measurable now. And it's growing.
Top comments (0)