Institutional World Model
Name a domain. This finds the occupations that actually staff it in the Department of Labor's O*NET taxonomy, then models each one task by task — what gets automated, what gets augmented, what stays human, and what that means for headcount.
Explore the whole map of work → · occupation graph →
Try · · ·
Only part of these occupations belongs to this domain.
Put your headcount against each role. The domain number is the sum of the role models — each one run on its own task list — so the two levels can't tell different stories.
How far past today's models. Above ~70% the human check becomes sampling rather than per-instance.
Share of exposed work where AI is actually deployed.
How much runs unsupervised rather than checked each time.
If the volume of work grows, freed capacity gets absorbed instead of cut.