Reading the tracker. Each bubble is one of the US occupations covered by the Anthropic Economic Index (AEI) with a full O*NET task profile. Y axis — realized AI usage share, importance-weighted mean of task-level penetration. X axis (C*) — structural potential, importance-weighted mean of each task's C* score, rescaled to 0 (pure physical) → 1 (fully autonomous). Bubble size = employment (BLS OES 2025). Colour = dominant C* class: Physical, Human-in-the-loop, Fully autonomous. The dashed line is the theoretical y = C* diagonal — full diffusion would put every occupation on it. The blue line is the actual, employment-weighted OLS fit — the gap between the two is how far realized adoption sits below the structural ceiling.

Reference. Verschuere & Cameron (2026), “The Two Axes That Sort AI Use: Task-Level Evidence from Platform Usage Logs” (SSRN 7408558) — the C* measure and its reproducibility across six language-model graders. Usage data: Anthropic Economic Index (Handa et al., 2025; arXiv:2503.04761), task-level penetration snapshot bundled with the paper’s replication package. Task inventory: O*NET 27.0. Employment: BLS OES 2025.