What does Claude's usage tell us?

Anthropic shared deep insights into usage on claude.ai over the course of 7 days. Half of this felt like a research paper and the other half felt like a product use report that someone could use to make claude better.

I took a few notes while reading

Claude handles 52M conversations per year

Programmers = early adopters

• This data is just Claude -- Imagine if it included Copilot/Cursor/etc.
• If programmers are a preview of the future, AI is going to be everywhere.
• Not a question of if AI gets adopted broadly, but how much people will use it.

Skills AI isn't touching (yet?)

• Are some skills missing from AI conversations because AI can't do them, or just because workers in those fields aren't using AI?
• Guide for "safe" jobs? Or a sign of arbitrage opportunities where AI hasn't been applied yet?

AI usage vs. wage

• Peaks in mid-to-high wage jobs (software engineers, copywriters, tutors, bioinformatics...)
• Lower usage in high-wage (doctors) & low-wage (physical labor) jobs—but is this because AI isn't useful there, or because these workers aren’t behind a computer all day?
• Are engineers the most productive workers? Or just the most AI-compatible?

Automation vs. Augmentation

• AI is mostly used to help people work better, not replace them.
• "Task iteration"—is this just directive automation waiting for better models?
• The 14.8% fully automatable work
  • There's already a chunk of work that could be automated fully but isn’t.
  • The real problem isn’t AI capability—it’s the missing infrastructure to harness it.
  • Expect this number to grow as better integrations emerge.