What do people do all day?
A recent Axios post repeats Dario Amodei’s warning that
AI could wipe out half of all entry-level white-collar jobs
and continues (in the staccato house style, so super strange) to dwell on the potential for disruption, without ever bothering to clarify: how many entry-level white-collar jobs are there, exactly?
I think most AI discourse, whether it’s hype or caution or both, suffers from a lack of clarity around what people actually do in the U.S. economy.
I’m not saying AI won’t be disruptive, and I’m definitely not saying I am some super expert on the structure of the U.S. labor force. I’m just saying, I don’t believe anybody writing these posts could answer basic questions about that labor force if you asked them.
Test yourself:
- About how many people in the U.S. have jobs classified as “computer and mathematical occupations”?
- About how many have “personal care and service occupations”?
- “Construction and extraction occupations”?
- “Sales and related occupations”?
And what is each category as a percentage of total U.S. employment?
Here are the answers.
This data is from 2023, only because I couldn’t find a comparable presentation on the BLS website for 2024 or 2025. Things have not changed hugely since then:
- Computer and mathematical: 6.5 million (4%)
- Personal care and service: 4.2 million (2%)
- Construction and extraction: 8.4 million (5%)
- Sales and related: 14.3 million (9%)
You look at the BLS data tables and there are, for sure, fat rows ripe for disruption. (Note that the modest total in farming doesn’t include actual farmworkers, who are not represented in this data; I suspect there are more distortions of this kind throughout the hands-on part of the economy.) The BLS doesn’t give us a figure for “entry-level white-collar jobs”—a doubly fuzzy category —
You might do the math differently and arrive at numbers that are indeed big and scary —
And maybe, continuing your investigation, you’ll want to cross-reference that data with this chart —
—in which the blue bars are as close a proxy for “maybe AI can do it” as you’ll find!
To the blog home page