Anthropic’s job market chart looks scary. Then you read the fine print.

You’ve probably seen the chart floating around. It’s from Anthropic’s latest report on AI and the labor market, and it’s designed to show two things: how much AI is already being used across different job categories (in red), and how much it could theoretically handle (in blue).

The red part is interesting but not shocking. The blue part, though, looks like a monster. It suggests that LLMs could theoretically perform at least 80 percent of tasks in categories ranging from Arts & Media to Legal, Business & Finance, and even Management. At a glance, it feels like Anthropic is saying AI is about to swallow half the economy.

I get why people are freaked out. That blue area is huge. But when you actually dig into how Anthropic arrived at those numbers, the picture gets a lot less dramatic.

The blue blob is not a prediction

Anthropic’s “theoretical capability” metric isn’t based on real-world performance or even rigorous testing. It’s a composite of educated guesses, expert surveys, and extrapolations from current model benchmarks. The company itself admits that the methodology is speculative. In fact, the report notes that these numbers represent a kind of upper bound — what AI might be able to do if everything goes perfectly, not what it will do anytime soon.

That’s a crucial distinction. A theoretical capability of 80 percent doesn’t mean AI can do 80 percent of the work in those fields today, or even next year. It means that, under ideal conditions and with significant advances, it could eventually handle that portion of tasks. That’s a very different claim from “AI is about to replace 80 percent of these jobs.”

The same chart also shows current observed exposure, which is much lower. For most categories, it’s under 10 percent. That’s the real story — AI is being used, but nowhere near the scale the blue blob implies.

What’s missing from the chart

The chart also lumps together “tasks” in a way that flattens important differences. A task like “draft a legal memo” might be 100 percent automatable on paper, but in practice, legal work involves judgment, client relationships, and context that LLMs can’t handle. The same goes for management and creative roles. The blue area assumes you can chop up jobs into discrete, AI-friendly tasks, but that’s not how most work actually functions.

There’s also the question of economic feasibility. Even if an LLM can do a task, it might not be cheaper or more reliable than a human, especially when you factor in oversight, error correction, and liability. A lot of the blue area is technically possible but practically pointless.

The takeaway

Anthropic’s chart is useful as a thought experiment, not as a forecast. It shows where AI could go if everything breaks right, but it doesn’t tell you when, how, or at what cost. The red area — actual usage — is a much better indicator of where things stand right now. And that area is still tiny.

So next time someone posts that chart with a caption like “AI is coming for all our jobs,” ask them how they define “theoretical capability.” The answer is probably less scary than the visual.