The official job map missed work people actually do: a language-model interviewer adds 79 tasks across six occupations, lifting holdout coverage from 78.6% to 89.3%
Synopsis
An occupational task catalogue for the US economy now has language models interview workers directly instead of relying only on fixed questionnaires. Standardized surveys like O*NET are slow to catch tasks that have just appeared, and human-led interviews are expensive to run at scale. The interviewer asks follow-up questions to clarify vague descriptions and confirms the resulting task bundles with the worker; on a holdout of 1,184 task reports from 57 workers, TaskNet's lists covered 89.3% of reported tasks on average across six occupations, against 78.6% for O*NET.
Interpretation
Researchers from Stanford, Mercor, the O*NET Resource Center and the National Bureau of Economic Research extended the task catalogue for six occupations beyond O*NET, adding 79 tasks that had no O*NET counterpart and that O*NET analysts validated, five of them created or reshaped as a direct result of AI adoption. O*NET collects task information mainly through standardized surveys, and a static questionnaire struggles to surface tasks that have just appeared, while interviews run by people are exploratory but expensive and hard to scale. On a holdout of 1,184 task reports from 57 workers whose interviews were not used to build the dataset, the catalogue's lists covered 89.3% of reported tasks on average across the six occupations, against 78.6% for O*NET itself.
A language model conducts the interview, asking follow-up questions when a description is vague and then confirming the resulting bundle of tasks with the worker. This is an AI interview rather than a fixed questionnaire: the next question depends on the answer, which keeps the exploratory character of interviews without assigning a person to every in-depth session. All 229 candidate task statements from the six occupations were reviewed by O*NET analysts, each met the center's standard for a valid task, and 21% were rewritten for departing from style guidelines or describing work mainly done by another occupation.
Perspective
The catalogue currently covers six occupations in the US economy, and the 89.3% coverage is an average over 1,184 holdout task reports from 57 workers; extending the interviews to more occupations and repeating them over time is the next step the release states.
During review, analysts flagged task statements that departed from O*NET style guidelines or described work mainly performed by another occupation, and 21% of tasks were rewritten. The five tasks created or reshaped by AI adoption come mostly from software engineering, with lawyers among the non-technical examples, and the billions of hours and roughly $140 billion in time-adjusted wages are estimates attached to the 79 new tasks.