Occupational Half-Life — How Many Years Until AI Halves a Role's Headcount

t½ = year headcount reaches 50% · N(t) = N₀ × (1 − D × adoption(t)) · D = 1 − (1 − β)^m

decays soon partial decline largely resists Eloundou et al. ↗

This is a projection, not a measurement. The spine is sourced — occupations, headcounts and wages from BLS OEWS (May 2021); task-exposure scores (β) from Eloundou et al., "GPTs are GPTs" (human-labelled). Two model assumptions sit on top, both made explicit and tunable below: adoption speed sets when, and displacement intensity (m) sets how hard task exposure converts into lost jobs — automating part of a role can shed more than that share of workers. A role whose modelled maximum loss D never reaches 50% has no half-life: it shows its projected decline (↓X%) instead, which is a decline, not safety. Searched roles are the same extrapolation surfaced on request, not extra evidence.

Adoption speed

Cautious Aggressive

Displacement intensity

Modest Severe

Find your occupation

Featured occupations	Exposure	Half-life

finisnostri.com · FN-004 · The Majevadia Complex