How Much of the Economy Runs Without People? Measuring the Automated Sector by Industry

Is a "fourth sector" of work done by AI without human steering visible in US data? We measure it as a share of task input for 19 industries from three exposure measures, Census firm adoption and one platform's automation shares. Exposure averages 0.30 on a 0–1 index. The performed share is not identified: no survey measures delegation depth, the share of exposed tasks adopting firms hand over. Across scenarios varying this depth (0.05 to 0.30) and four other assumptions, it is 0.1% to 2.4% of task input (central 0.7%); at full adoption and delegation 9.3% to 21.1%. Exposure predicts where firms adopt (rank correlation 0.86; 0.85 excluding usage-built measures); none with adoption speed was detected. One survey question could close this gap. In an assumed firm-level scenario, shared AI workflows could free and reuse time worth USD 430 per employee-year from ad hoc use (USD 269 from regular use), before adoption costs. Working paper; preliminary, comments welcome. The record holds the paper, the Online Appendix and the replication package (code under the MIT licence; raw data are not redistributed). JEL codes: O33, O41, J24, E25, L80, O14.