The real AI divide isn't between people who use AI and people who don't. It's between people who use AI well and people who are just getting started.
That's the core finding from Anthropic, shared at the Axios AI+DC Summit on March 25 by Peter McCrory, the company's head of economics. Anthropic has been studying Claude usage patterns across 20 higher-income countries, and the data tells a consistent story: experienced users are pulling further ahead, and the gap is hardening.
The Inequality Nobody Predicted
Most AI job-loss predictions follow a simple script: AI replaces workers, unemployment spikes, society adjusts. The Anthropic data suggests something more subtle and possibly more damaging. AI isn't replacing anyone's job yet. But skilled users are getting measurably better at collaborating with Claude to complete high-value work, while newer users plateau at basic tasks.
McCrory, who co-authored a paper with Anthropic economist Maxim Massenkoff, frames this as laying groundwork before AI effects have fully emerged. The idea is to identify economic disruption patterns early, while there's still time to address them.
This tracks with what anyone who's spent serious time with AI tools already knows. There's a steep learning curve between "ask ChatGPT a question" and "use Claude to draft a full regulatory analysis with source citations." That gap represents real economic value, and right now it's accruing disproportionately to people who were already tech-savvy.
Skills Gap or Class Gap?
The uncomfortable implication: AI fluency is following the same distribution pattern as every other valuable skill in the economy. People with more education, higher incomes, and white-collar jobs are adopting AI faster and using it for more sophisticated tasks. People without those advantages are either not using AI at all or using it at a surface level.
Anthropics's data shows this pattern persisting across all 20 countries they studied. This isn't a temporary adoption curve that flattens as tools get easier. The gap is widening.
Granted, this research comes from a company with a direct financial interest in making AI seem important. Anthropic sells Claude subscriptions, and "you need to get better at AI" is a convenient marketing message. But the underlying observation matches what hiring managers and productivity consultants are reporting independently: AI proficiency is becoming a real differentiator in job performance, and it's not evenly distributed.
What This Actually Means for You
The practical takeaway isn't complicated. People who invest time learning to use AI tools effectively - not just prompting, but understanding what these tools can and can't do, building workflows around them, knowing when to trust the output and when to verify - are gaining a compounding advantage. Every month of serious use makes the next month more productive.
The policy question is harder. If AI skills are becoming as economically important as Anthropic suggests, then access to quality AI tools and training isn't just a nice-to-have. It's a workforce development issue on par with computer literacy in the 1990s. The difference is that this transition is happening much faster.