OBSERVED IN REAL USE · Anthropic 2026
18%
of this role’s work is already showing up in real Claude usage (Anthropic Economic Index).
The model already does most of this before its first coffee.
98% of this role’s O*NET tasks are within reach of today’s AI. That is the core-weighted exposure score from Eloundou et al. 2023 (“GPTs are GPTs”). It measures a capability ceiling, not a headcount forecast. Deeply cooked. Almost the entire task list is something a model can already attempt.
WHAT AI CAN ALREADY DO
- Mark copy to indicate and correct errors in type, arrangement, grammar, punctuation, or spelling, using standard printers' marks
- Read corrected copies or proofs to ensure that all corrections have been made
- Correct or record omissions, errors, or inconsistencies found
- Compare information or figures on one record against same data on other records, or with original copy, to detect errors
- Route proofs with marked corrections to authors, editors, typists, or typesetters for correction or reprinting
WHAT IT STILL CAN’T
- Almost nothing here is clearly out of reach. That’s the warning in this one.
THE HONEST PART. A percentage is not a pink slip. High exposure usually means a role shrinks and shifts toward judgment, direction and responsibility: the parts a model can’t sign its name to. Exposure ≠ displacement. Breathe.
"My job is 98% cooked. What’s yours?"
SOURCES: O*NET 30.3 occupational tasks · Eloundou et al. 2023 (“GPTs are GPTs”,
arXiv:2303.10130) · Anthropic Economic Index 2026 (CC-BY) |
how this is calculated |
last updated 2026-07-16