Anthropic just published the most detailed look yet at how AI is reshaping work. The June 2026 Anthropic Economic Index report, based on millions of Claude conversations and a survey of 9,700 users, reveals a workforce that knows the ground is shifting under its feet.
Over one third of respondents expect AI to handle most or nearly all of their work tasks within 12 months. The people who delegate the most to AI are the most optimistic about their careers. And the youngest workers are the most afraid of losing their jobs.
These are not predictions from pundits. This is data from the company that builds one of the world's most capable AI models, drawn from real usage patterns and direct survey responses. (Anthropic Economic Index)
What the Economic Index Actually Measures
Anthropic launched the Economic Index in early 2026 to track how people use Claude across every US state and hundreds of occupations. The June 2026 report, titled "Cadences," is the fourth installment and the most ambitious. It introduces three new data sources: hourly usage sampling, an output classifier that labels what each conversation produces, and the Economic Index Survey. (Anthropic, June 2026 Report)
The dataset is public. Anthropic publishes it on HuggingFace, and the full methodology is in an open appendix. This is not a black-box study.
Finding 1: Your Workweek Is Etched Into AI Usage
Claude usage mirrors the workweek with striking precision. On weekdays, roughly 35% of conversations are personal. On weekends, that jumps to nearly 50%. Business correspondence, marketing copy, and slide decks give way to emotional support, medical questions, and investment advice.
The hourly data is even more revealing. People ask for news at 7 a.m. Business correspondence peaks at 10 to 11 a.m. Recipe requests spike to 2.3 times the average at 6 p.m. Sleep advice peaks around 5 a.m. Tax-related conversations surged to 8 times the average on April 14, the day before the US filing deadline, then dropped sharply on April 16.
On weekends, when people do turn to Claude for work, the tasks skew toward higher-wage occupations. Tasks related to jobs in the bottom two wage quartiles, like telemarketing and clerical work, shrink. This pattern holds even when you remove computer and mathematical occupations entirely. People in higher-paying jobs work more outside traditional hours, and they bring AI with them.
Finding 2: Higher-Wage Work Costs More Compute
Conversations mapped to higher-wage occupations consume more tokens. A lot more. Marketing managers earn roughly twice what editors earn ($80 vs. $37 per hour), and their Claude conversations consume approximately 2.5 times as many tokens.
This is not just because higher-wage work is more complicated. It is also because people in those roles produce more with Claude (1.34 times as much output per turn), engage in more turns (1.53 times as many), and enable extended thinking more often (34% of conversations versus 31%).
The key insight: more Claude production does not mean less human production. Users who get more from Claude also put more in. The pattern looks like labor augmentation, not labor displacement. At least for now.
About 44% of the wage gradient in token consumption is explained by output mix. Higher-wage occupations simply produce more compute-intensive artifacts. Building an app consumes more than 3 times the tokens of the median conversation. A typical explanation uses about a fifth.
Finding 3: Claude Code Changes Everything About How People Work
The report introduces a 1-to-5 autonomy scale measuring how much decision-making power Claude has in a conversation. Across 26 of 31 output types, Claude Code sessions show higher autonomy than chat or Cowork sessions.
The difference is stark. Blog posts and articles illustrate it. On chat, the median blog post conversation involves 13 rounds of back-and-forth. On Claude Code, the median session contains a single human prompt. The person writes one instruction, and Claude builds the entire thing.
Approximately two thirds of the autonomy gap comes from the same tasks being executed with more delegation on Claude Code. The remaining third comes from a different mix of output types.
This matters because Claude Code runs on the most capable models far more often (54% on Opus, versus 10% of chat sessions). But even when you compare conversations served by the same model, the autonomy gap persists. Sonnet sessions on Claude Code still show 0.26 points more autonomy than Sonnet sessions on chat. The product, not just the model, determines how much people delegate.
Finding 4: 35% Expect AI to Do Most of Their Job Next Year
This is the headline number from the Economic Index Survey. Close to 6 in 10 respondents chose a higher capability band for next year than for today. Over one third expect AI to be able to do most or nearly all of their work tasks within 12 months.
Here is the twist: the people who delegate the most to Claude are the most optimistic about their own careers. Across six dimensions of job quality (pay, job security, ability to find a job, meaning, autonomy, and human interaction), people with a higher share of automated sessions feel more positive about AI's effect on their work. The largest effects are on expectations about future pay and job-finding ability.
Large majorities report productivity gains: 86% say AI makes them faster, 82% say it expands their scope, 69% say it improves quality. And 57% say AI has made their skills more valuable. This rises with automation share. Heavier delegators report learning at the same rate as everyone else.
The study authors acknowledge a selection problem. Maybe the people most enthusiastic about AI are also the most willing to delegate. They cannot rule it out. But the pattern holds even when controlling for user tenure on Claude, a proxy for early adoption enthusiasm.
Finding 5: Young Workers Are Scared
Early-career workers report that AI can do the highest share of their work and express the most concern about job loss. Respondents were especially worried about job loss for junior colleagues: over one third said the probability of a junior colleague losing their job in the next year was above 60%.
This lines up with Anthropic's earlier labor market research. A separate study using Current Population Survey data found a 14% drop in the job-finding rate for workers aged 22 to 25 in the most AI-exposed occupations since ChatGPT launched. The effect is just barely statistically significant, but the direction is consistent. (Anthropic, Labor Market Impacts)
The mechanism appears to be slowed hiring, not increased layoffs. Young workers are not being fired. They are simply not being hired into the most exposed roles at the same rate.
Finding 6: The Global Divide Is Real
Per-capita Claude usage varies enormously across countries. Australia tops the index at 6.4, followed by Singapore at 5.81, Switzerland at 5.02, and Luxembourg at 4.85. The United States sits at 3.87. India, despite being the second-largest market by total volume, scores 0.30 on the per-capita index. Bangladesh scores 0.11.
Workers in lower-income countries report that AI can do a larger share of their tasks. This is consistent with the IMF's analysis that while advanced economies face broader AI exposure, workers in developing nations may have less access to the complementary skills and infrastructure that turn AI from a replacement into an augmentation tool. (IMF, AI and the Future of Work)
In Anthropic's earlier work, they documented that lower-income economies use Claude in more automated ways, even after adjusting for differences in task mix. The pattern is clear: where AI substitutes for tasks rather than augmenting them, the risks of displacement are higher.
Finding 7: Women Use AI Differently
Women make up only 12% of the survey's linked respondent sample. Even after accounting for occupational differences, women use Claude differently from men. Their share of sessions in Claude Code is 0.24 standard deviations lower. Their automation share is 0.33 standard deviations lower. Instead, women tend to use Claude more iteratively, logging more active time on chat.
The study does not speculate on why. But the pattern is consistent across occupational controls. Women collaborate with AI. Men delegate to it.
The Bigger Picture: No Unemployment Spike Yet
Despite all these signals, Anthropic's labor market research finds no systematic increase in unemployment for workers in highly AI-exposed occupations since late 2022. The difference-in-differences estimate is small and statistically insignificant.
This does not mean AI is not affecting jobs. It means the effect has not yet shown up in aggregate unemployment data. The BLS projects slower growth for occupations with higher observed exposure: for every 10 percentage point increase in AI coverage, projected employment growth drops by 0.6 percentage points. But projections are not outcomes.
The most exposed occupations include computer programmers (75% task coverage), customer service representatives, and data entry keyers (67%). The least exposed include cooks, motorcycle mechanics, and bartenders, where 30% of workers have zero AI coverage.
What People Actually Hope For
The survey ends with an open-ended question: what do you hope an economy shaped by AI looks like in ten years? The top five themes:
- AI augmentation of meaningful work (over half of respondents). Collaboration, not replacement.
- Automation of drudgery (just over half). Freeing time for what matters.
- Shared prosperity (about one third). The gains should not concentrate at the top.
- New industries and opportunities. Hope that AI creates, not just destroys.
- Human connection preserved. The relational and interpersonal parts of work matter.
These are not the hopes of people who have given up. They are the hopes of people who use AI every day and want it to make their work better, not make their work irrelevant.
What This Means for You
Three takeaways for anyone watching this space:
Learn to delegate, or learn to collaborate. The data shows that people who delegate more to AI are more optimistic about their careers. This could be selection bias, but the consistency of the finding across multiple controls suggests something real. The workers thriving with AI are not avoiding it. They are leaning in.
Junior roles are the canary. If AI displaces work, it will show up first in entry-level hiring. The 14% drop in job-finding rates for young workers in exposed occupations is the strongest early signal in the data. If you are early in your career, focus on skills that require judgment, context, and human relationships, the things the survey respondents themselves say AI cannot do.
Where you live matters. AI augments in high-income countries and substitutes in low-income ones. If your economy lacks the infrastructure, training, and institutional support to turn AI into a productivity multiplier, the same technology that creates opportunity elsewhere can eliminate it where you are.
Anthropic has done something rare here. They built a public dataset, published their methodology, and shared findings that complicate their own commercial narrative. The data does not say AI is replacing everyone. It also does not say AI is harmless. It says the impact is uneven, early, and accelerating.
The full report is at anthropic.com/research/economic-index-june-2026-report. The dataset is on HuggingFace. The interactive dashboard is at anthropic.com/economic-index.
Sources:
Top comments (0)