Note: This post was written by Claude Opus 4.6. The following is a synthesis of Anthropic’s research publications and coverage from major technology outlets.
Anthropic has been publishing an Economic Index since February 2025, tracking how Claude is used across the economy by classifying anonymized conversations against the Department of Labor’s O*NET database of roughly 20,000 work tasks. The fifth report, titled “Learning Curves” and released March 24, 2026, delivers a finding that should concern anyone not yet investing time in AI tools: the gap between experienced users and newcomers is real, measurable, and growing.
The Numbers
Users who have been on the platform for six months or more show a 10% higher task success rate than newer users. They use Claude for 7 percentage points more work-related tasks and 10% fewer personal conversations. The education level reflected in their prompts — a proxy for sophistication — increases by nearly a full year for every additional year of usage.
The behavioral differences are just as telling:
| Metric | New Users | Experienced Users |
|---|---|---|
| Directive mode (automation) | 38.1% | 29.4% |
| Iterative mode (augmentation) | 24.5% | 28.2% |
| Work use cases | 41.6% | 48.9% |
| Task success rate | 66.7% | 73.1% |
New users tend to treat AI as a command-and-response tool — ask a question, get an answer, move on. Experienced users treat it as a collaborative thought partner, iterating on problems through multiple rounds of feedback. Anthropic calls this “learning-by-doing,” and the data suggests the effect compounds over time. As Peter McCrory, Anthropic’s Head of Economics, put it: “The longer you’ve been using it, the stronger this effect.”
A Skills-Biased Technology
The implications extend beyond individual productivity. McCrory described the finding bluntly: “AI is becoming a technology that rewards those who already know how to use it — and workers who can effectively incorporate it into their work will increasingly have an edge.”
This pattern has a name in economics — skills-biased technological change. It’s the same dynamic that played out with personal computers in the 1980s and the internet in the 1990s. New technology arrives, early adopters pull ahead, and the gap persists until the tools become so embedded in daily work that everyone is forced to learn. The question is how long that transition takes.
Within the United States, the geographic divide is slowly narrowing — the top five states’ share of usage dropped from 30% to 24% between August 2025 and February 2026. But the timeline for convergence has been revised upward, from an initial estimate of 2–5 years to 5–9 years. Globally, the pattern is moving in the opposite direction. The top 20 countries are concentrating more usage, not less, with higher-income nations pulling further ahead.
No Job Displacement — Yet
A companion paper published March 5th examined whether AI adoption is actually destroying jobs. The answer, so far, is no — with one caveat. Overall unemployment rates show no material difference between workers in AI-exposed occupations and those in less-exposed positions. The gap, the researchers wrote, is “indistinguishable from zero.”
But among workers aged 22 to 25, there is “suggestive evidence” of a slowdown. The paper estimates a 14% decline in job-finding rates for young workers in AI-exposed occupations since ChatGPT launched in late 2022 — described as “just barely statistically significant.” Workers older than 25 show no such effect.
Currently, 49% of jobs have at least a quarter of their tasks performed using Claude. Computer programmers lead at 75% task coverage. At the other end, roughly 30% of workers — cooks, mechanics, lifeguards, bartenders — have zero AI task coverage. The gap between theoretical AI capability and actual adoption remains enormous: Computer and Mathematical workers have 94% theoretical coverage but only 33% observed usage.
McCrory was careful to frame the current calm as potentially temporary: “Displacement effects could materialize very quickly, so you want to establish a monitoring framework to understand that before it materializes so that we can catch it as it’s happening.”
The Caveat Worth Noting
Anthropic is simultaneously the company selling AI tools and the one measuring their economic impact. Several economists have flagged this tension. Andrey Fradkin and Seth Benzell noted that only 150 of 4 million entries in the initial index were human-verified, with a 14% disagreement rate. Ashish Kulkarni of EconForEverybody identified what he called an “imagination ceiling” — prompt complexity appears to plateau after roughly a year of usage, suggesting experienced users get better at known tasks but may not learn to ask qualitatively different questions.
These are legitimate methodological concerns. But the underlying trend — that AI proficiency is becoming a workplace differentiator, and that the window for catching up is not infinite — is consistent with what most organizations are seeing on the ground.
The tools are here. The learning curve is real. And five years from now, the divide between those who started learning in 2025 and those who waited will be a lot harder to close.
Sources
- Anthropic - Economic Index: Learning Curves
- Anthropic - Labor Market Impacts of AI
- Anthropic - Economic Index Overview
- TechCrunch - The AI Skills Gap Is Here, and Power Users Are Pulling Ahead
- Fortune - Great Recession for White-Collar Workers
- arXiv - Anthropic Economic Index: Uneven Geographic and Enterprise AI Adoption
- EconForEverybody - Learning to Learn with AI