AI at Work Three indices: use, jobs and training
Sources

Is AI changing American work yet? No single dataset or study can say. This page brings together public data, published research and the main ideas about how technology changes work, in three indices that update as new releases come out. Any one source is a glimpse through fog; read together, they show more.

As of the October 2026 update, 23% of US wages are paid for tasks people also bring to AI models (Claude data). Pay hasn't changed more than usual, and job ads turned away from tasks AI can do from 2020 to 2022, before ChatGPT, with no further drop at its release. And 42% of new degrees are in fields that lead to the jobs most exposed to language models.

The city (Jobs) whether AI has changed pay, jobs or hiring The bridge (Use) where AI could change work first The binoculars (Training) how the next generation of workers is adjusting

Three views from one point

Choose the bridge, the city or the binoculars to open that index.

Figures from the latest build.

A series that grows with every release

The page always shows the most recent data that exist, and every series grows longer with time. Each source gets updated when new releases appear, adding a new point to our series.

A natural next layer would be data on what firms spend on AI, such as the Ramp AI Index, since it measures adoption where firms actually pay.

The Census Bureau publishes firm adoption every two weeks; OpenAI Signals and the Current Population Survey come out monthly; the Anthropic Economic Index updates with each release; and pay, jobs, degrees and enrollment update once a year.

Sources

Use (the bridge)

Anthropic Economic Index, use of Claude by O*NET task, with the split between automation and augmentation, releases from September 2025 to the June 2026 monthly file. OpenAI Signals v2.0, work-related messages by work activity, June 2026, shown beside Claude on the skill-group card; each platform sees only its own users, so the two are never averaged. BTOS, the share of US firms using AI, Census Bureau, biweekly from November 2025 to September 2026, with its published standard errors. OEWS, wages and employment by occupation, BLS, May 2025.

Jobs (the city)

OEWS, wages and employment by occupation, BLS, May 2012 to May 2025. O*NET 30.3, tasks, ratings and skills, Department of Labor. NLx job ads, the tasks named in the ads of the National Labor Exchange, monthly from September 2015 to September 2025, in the aggregates of Meisenbacher, Nestorov and Norlander. Occupation crosswalks from the Census Bureau, NCES and BLS.

Training (the binoculars)

IPEDS completions by field and institution, NCES, 2023 to 2025. National Student Clearinghouse, undergraduate enrollment at four-year institutions by field family, fall 2019 to fall 2024. The NCES CIP-to-SOC crosswalk, which links each field of study to the occupations it is meant to lead to. ACS 2024, field of degree against occupation, computed as a check on the crosswalk. CPS basic monthly, Census Bureau, January 2022 to August 2026. Exposure is the task measure of Eloundou, Manning, Mishkin and Rock (2024), with the ability-based measure of Felten, Raj and Seamans (2021) as the alternative on the degree and enrollment cards.

About the drawing

The drawing is the view from the lighthouse at Point Bonita, at the mouth of the Golden Gate, looking across to the bridge and the city and, through the coin-operated binoculars, south down the peninsula to the campuses. The weather is the latest observation at San Francisco International Airport (station KSFO), published by the National Weather Service: its description and visibility set the sky and the fog, its wind sets how fast and which way they drift, and rain falls when the station reports it. The sun and the moon are placed for the time in San Francisco from the standard formulas for their positions, and the moon is drawn at its phase. The boats, planes, trains, whales and gulls are invented.