Yesterday Anthropic published a labor market report. Worth reading, but this chart alone tells most of the story.

Blue is theoretical AI capability by occupation. Red is what people actually use Claude for in professional settings.

For Computer & Math jobs, the blue reaches 94%. The red sits at 33%.

Everyone arguing about AI displacement is arguing about the blue. The red is what’s actually happening.

And right now the red is small. Not because the capability isn’t there. Because capability and deployment are different curves, with different friction, different timelines, different blockers.

The report flags some of those: legal constraints, verification requirements, specific software integrations. Anthropic’s own example is prescription authorization. Theoretically fully automatable. Observed usage: zero.

So when someone tells you “AI will replace X” they’re reading the blue chart. When someone says “I’ve been doing this job for two years and AI hasn’t changed anything,” they’re reading the red one.

Both people are probably right. They’re just measuring different things.

ps: link in the comments

Media