Where AI actually lands at work.
Most coverage about AI and work focuses on which jobs will change. This map shows something more specific: which occupational categories are already seeing AI use, and how far that usage falls short of what the tasks theoretically allow. The bigger the cell, the more exposure. The colour shows how much of the potential is already active.
Data: Anthropic Economic Index, Feb–Mar 2026 report. Retrieved June 2026. Theoretical capability for 14 of 22 categories estimated from Eloundou et al. 2023 (see source note below).
Cell size = share of tasks in each category where Claude is actively being used today.
The bigger the cell, the more of that category's work involves tasks where AI is actively being used. The toggle above switches cell size between observed exposure (what is happening now) and theoretical capability (what the tasks would allow).
Colour shows how much of the theoretical potential is already in active use. Green cells are categories where AI is already deeply embedded in the actual work relative to what the tasks allow. Coral cells are categories with significant theoretical potential that has not yet appeared in observed usage. The colour is a usage-intensity signal, not an assessment of whether AI in that category is positive or negative for workers.
What the map shows is where AI shows up, not which jobs are at risk. High exposure means the tasks in a category are ones AI can assist with. It does not determine whether workers will be displaced or augmented. One separate site-wide figure from the same dataset: across all interactions in the Anthropic Economic Index data, about 57% follow an augmentation pattern (a person directing the tool) and 43% follow an automation pattern. That split is an aggregate across all categories and does not map to the per-category colour shown here.
Primary source: Anthropic Economic Index, labor market impacts data, Feb–Mar 2026 (Hugging Face, CC-BY). Observed exposure for the eight highest-exposure categories uses employment-weighted averages from the published report. The remaining 14 categories use simple averages from the released job_exposure.csv file, which may differ from employment-weighted figures. Theoretical capability for eight categories (Computer and Mathematical, Office and Administrative Support, Business and Financial, Management, Architecture and Engineering, Legal, Arts/Design/Media, Sales) comes from the AEI report directly. The remaining 14 use estimates from Eloundou et al. 2023 (Science, 2024), a peer-reviewed O*NET task analysis using GPT-4 that should be read as a lower bound. The aggregate automation vs augmentation split (57% / 43%) is from the Anthropic Economic Index primitives report, January 2026.
The dataset is based on real Claude.ai conversations mapped to occupational categories. It skews toward early, white-collar users. It captures where AI use is currently most visible, not necessarily where it will land over time. The ILO (World Employment and Social Outlook 2024) and Yale's Budget Lab separately identify broadly the same categories as highest-exposure, which gives these figures additional weight.