Other | Current estimate | Data through Jul 2026
Observed AI Use at Work
An estimated 35% of US workers have AI observed doing some part of their actual work. This is the measured counterpart to exposure: platform telemetry and nationally representative worker surveys, rather than capability mapping. The number is sensitive to where the threshold sits. Requiring AI to touch at least a quarter of a job's tasks yields 49%; counting any observed coverage at all yields about 70%. Reach has run well ahead of depth: AI shows up somewhere in occupations covering 88% of US employment, but the median occupation with any use runs it on only 21% of its tasks.
This is observed data from real-world surveys and measurements, not a prediction. See the full methodology for details on weighting, source validity, and recency bias.
Indicators Over Time
The chart below tracks how this estimate has shifted over time as new research and data emerge. Every source is color-coded by evidence quality; use the tiers below to filter what appears on the chart and in the weighted average above.
Filter by evidence tiers
Note: Read the spread here as a disagreement about definitions, not about facts. Each source draws the line for “using AI” in a different place, and where you draw that line does most of the work. Anthropic's 70% counts a worker if any of their tasks show up at all in professional AI traffic. Apollo's 3.7% counts a worker only if AI covers at least half their tasks. Hartley et al.'s 38% skips telemetry entirely and asks workers whether they use AI. All three can be correct at once. Because of that, the average across them is not a meaningful estimate of how many people use AI at work — treat the individual points as answers to three different questions. Breadth is also not depth: Google's ATLAS observes Gemini somewhere inside occupations covering 88% of US employment, but the typical occupation with any use runs AI on only 21% of its tasks. That 88% measures how far AI has reached across the economy, not how many workers use it, which is why it sits below the chart rather than on it. For what AI could do rather than what it is doing, see US Workforce AI Exposure.
Directional research signals
Studies that point in a clear direction but give no single number to chart — e.g. “entry-level hiring fell” or “no measurable displacement detected.” They are not counted in the average above. Stacked blocks show net evidence per month; positive and negative signals cancel. Hover any column to see the studies.
Each data point is from a different source. Dots are color-coded by evidence tier. Click any dot to jump to its source.
Task Visualizer
What parts of your job will be cheaper to do with AI?
See which of your tasks face cost pressure from AI first.
Full Economy Picture
AI and the US Economy
Automation impact by occupation and income tier.
Sources (7)
NY Fed: 61% of service firms use AI but median adopter has only 17% of staff on it
Show excerptHide excerpt
Third annual AI module in the NY Fed's regional business surveys: "61 percent of service firms reported using AI this year, up from 40 percent last year and 25 percent in 2024," while among manufacturers "51 percent reported using AI as part of their business processes, roughly double the 26 percent from last year and triple the 16 percent in 2024." The bar is stricter than most adoption surveys — firms using AI "exclusively as an information search tool but nothing else" are coded as non-users. Adoption is wide but shallow: "Three-quarters of service firms and more than 90 percent of manufacturers characterize their AI investments as minimal to modest," and among adopters "the median share of workers using it was just 17 percent for service firms and 7 percent for manufacturers." On labor, "only 4 percent of service firms reported laying off workers" because of AI (up from 1 percent in 2025), "about 15 percent of service firms said they had hired fewer workers than they would have if not for AI use," and "about 13 percent of service firms said they had hired more workers due to AI" — while "just over a third of service firms and more than 20 percent of manufacturing firms report retraining workers." Recorded as overlays: the survey covers New York State and northern New Jersey only, so it is not comparable to the national Census BTOS series, and the workforce shares are of AI-adopting firms rather than of jobs or workers. The authors caution that "AI technology and its applications are still evolving rapidly, and these patterns could shift as adoption matures."
Show excerptHide excerpt
5.8 million workers in the U.S. are currently working in high exposure occupations, which represents about 3.7% of the U.S. labor force. High exposure is defined as an Anthropic Economic Index observed-usage score of 0.5 or above, a stricter threshold than Anthropic's own 25%-of-tasks cut (49%) or any-observed-coverage cut (70%).
Google ATLAS: median occupation uses AI for only 21% of tasks — broad but shallow
Show excerptHide excerpt
Google's AI & Economy ATLAS v1.0 (Iscenko, Strand, Chen, Imas, Manyika et al., Google/Google DeepMind, Jul 23 2026; reviewed by Diane Coyle and David Autor). 15M de-identified interactions across Gemini App, AI Mode, and Gemini API (Apr 6-19, 2026), mapped to 800+ occupations, 4,000 O*NET tasks, 300 ATUS activities, 150 countries, 140 languages. Work: 'AI adoption spans occupations covering just above 88% of US employment' (68% of detailed occupations) but 'penetration remains shallow' — 'AI is used for only 21% of total tasks in the median occupation with any AI use'; only 3% of occupations show usage for >75% of tasks. 'Attempts to automate tasks end-to-end represent less than 10% of AI conversations in non-routine cognitive work'; >25% for routine cognitive work. Non-routine cognitive tasks = 35% of O*NET universe but 65% of work interactions. Wages: 'a 1% increase in an occupation's median earnings is associated with a more than 2.5% increase in AI usage intensity'; Gemini-weighted median salary $82,919 vs $62,252 employment-weighted national median. Home: 86% of conversational use is non-work; government/civic queries over-represented ~20x vs time spent; household time-savings valued at $14.9B-$149B/yr (0.5-5% scenarios). Global: 1% GDP/capita ↑ → 0.9% usage ↑; English only ~1/3 of conversations across 143 languages.
OpenAI: 66pp capability overhang — 90% theoretical vs 23.8% realized exposure in high-risk jobs
Show excerptHide excerpt
Jobs at high automation risk: gap 66.2 pp, 23.8% realized, 90.0% theoretical. Jobs that grow with AI: gap 49.7 pp, 22.7% realized, 72.4% theoretical. Jobs that will reorganize: gap 52.3 pp, 14.9% realized, 67.1% theoretical. Jobs with less immediate change: gap 21.0 pp, 6.4% realized, 27.4% theoretical. Across every job category, current usage lags behind the possible. Exposure alone is a weak predictor of immediate labor market pressure.
Fed SBU: 78% of US labor force works at an AI-adopting firm (employment-weighted)
Show excerptHide excerpt
The SBU estimates an employment-weighted firm AI adoption rate of around 78 percent and an LLM adoption rate of about 54 percent. In this context, employment weighting approximates the share of the labor force working at firms that have adopted AI.
Show excerptHide excerpt
At the bottom end, 30% of workers have zero coverage, as their tasks appeared too infrequently in our data to meet the minimum threshold.
Show excerptHide excerpt
LLM adoption among U.S. workers increased from 30.1% to 38.3% between December 2024 and December 2025. Small effects on wages in exposed occupations; no significant effects on job openings or total jobs.
Know a study we’re missing? Suggest a source for this prediction.