Source: ai-research/anthropic-economic-index-june-2026-cadences.md, raw/x-account-anthropicai-2070528971687755796.md
Publisher: Anthropic (Economic Research)
URL: https://www.anthropic.com/research/economic-index-june-2026-report
Published: 2026-06-26
The June 2026 edition of Anthropic’s recurring Economic Index series, titled “Cadences,” marks a methodology shift driven by how Claude usage has changed. A year ago most usage was a chat between a user and an assistant; with the rapid growth of Claude Code and Cowork, sessions now increasingly consist of long-running agentic tasks, so chat transcripts no longer fully capture how people use AI. Anthropic adapted its data pipeline in two ways: a new continuous Cadences telemetry method that samples conversations daily (enabling daily/hourly usage analysis), and the launch of the Anthropic Economic Index Survey to capture impact outside user sessions. The headline survey result inverts a common worry: the people who use Claude in the most automated way both expect AI to take on more of their work and feel the most optimistic about it.
Key Takeaways
- Usage outgrew the chat transcript. Because Claude Code + Cowork sessions are increasingly long-running agentic tasks rather than back-and-forth chats, transcripts alone no longer characterize usage — prompting the adapted pipeline behind this report.
- Cadences = continuous telemetry. A new privacy-preserving method that continuously samples a slice of conversations every day, in contrast to the seven-day samples every prior Economic Index report drew on. It exposes daily and hourly usage rhythms worldwide.
- External-world rhythms shape usage (Chapter 1): work queries subside on weekends (less so in the highest-paid occupations); people ask for the news in the morning; sleep-advice requests peak ~5 a.m.; tax requests surge around filing deadlines.
- The Economic Index Survey (Chapter 3): launched April 2026; ~9,700 linked respondents; usage linked mid-May to early June via privacy-preserving methods. Headline: the most-automated users expect AI to take on more of their tasks next year, yet are the most optimistic — anticipating positive impacts on pay, job security, and meaning.
- Tenure correlates with work use and success (secondary): users on Claude 6+ months are ~7 pp more likely to use it for work;
5 pp more likely to have a successful conversation in simple comparisons (3-4 pp after controls). - Partial extract. Chapter 2 of the report is not fully captured here; the full report, appendices, and PDF are at the URL above.
Cadences (new telemetry)
The defining methodological change in this edition. Prior Economic Index reports characterized usage from seven-day samples. Cadences instead continuously samples a slice of conversations every day using a privacy-preserving method, which lets Anthropic study daily and hourly usage patterns and capture the ebbs and flows in work patterns around the world.
The motivation is structural, not cosmetic: a year ago most Claude usage was a conversation between a user and an assistant, but with Claude Code and Cowork, sessions now increasingly consist of long-running agentic tasks. Chat transcripts no longer fully capture how people use AI — so Anthropic adapted the data pipeline (Cadences telemetry plus the new survey, below) to keep measuring economic signal as the shape of usage changes.
Usage rhythms (Chapter 1)
Chapter 1 examines how external-world rhythms shape Claude usage. Patterns surfaced by the daily/hourly sampling:
- Work-related queries subside on the weekend — though less dramatically in the most highly paid occupations.
- People tend to ask for the news in the morning.
- Sleep-advice requests peak around 5 a.m.
- Tax-related requests surge around filing deadlines.
Anthropic’s own announcement thread adds further hour-by-hour color (secondary): recipe prompts cluster in the evenings, and gardening interest stays roughly stable across the day.^[Per @AnthropicAI’s launch thread, marked secondary — the X-thread synthesis, not yet verified line-by-line against the primary report.]
The Economic Index Survey (Chapter 3)
A new instrument designed to capture AI’s impact outside user sessions — perceptions of how AI is changing work and opportunity.
Method:
- Survey launched April 2026; responses linked to Claude usage data from mid-May to early June using privacy-preserving methods.
- To characterize each respondent’s usage, Anthropic randomly samples up to 20 sessions per person within the window, across Claude.ai, Cowork, and Claude Code (so the mix reflects each person’s typical usage across surfaces).
- Respondents with fewer than five sessions are excluded to reduce sampling noise.
- Final linked sample: ~9,700 respondents.
Headline finding: expectations and experiences vary systematically with how people use Claude. People who use Claude in the most automated way expect AI to take on more of their tasks in the next year, yet feel the most optimistic about what that means for their work — anticipating positive impacts on pay, job security, and meaning.
Anthropic’s launch thread sharpens the picture (secondary): nearly half of respondents expect major changes to their job responsibilities within 12 months, the heaviest delegators are the most optimistic about their own pay and job security, and respondents express more concern for their junior coworkers than for themselves.^[Per @AnthropicAI’s launch thread, marked secondary — not yet verified line-by-line against the primary report.]
Corroborating coverage (secondary)
Per EdTech Innovation Hub’s summary of this report: users on Claude 6+ months are about 7 percentage points more likely to use it for work (and less for personal queries), with prompts associated with higher education and a wider task spread. Higher-tenure users are about 5 pp more likely to have a successful conversation in simple comparisons, narrowing to ~3-4 pp after controlling for task type, request cluster, country, model, and use case.
What it means for a marketing / agency reader
- Time-of-day and day-of-week matter for AI touchpoints. If client audiences mirror these cadences — news in the morning, work tapering on weekends, recipes in the evening — then the windows when people are in an AI tool asking questions are predictable, which bears on when AI-surfaced content (and AI-assisted support) gets used.
- “Automation-forward” users are the optimists, not the anxious. The clients and team members handing the most work to Claude are the ones expecting (and welcoming) AI taking on more of their tasks — useful framing when positioning automation to a skeptical buyer who assumes heavy AI use breeds job-security fear.
- Tenure compounds. The 6+-month-use lift in both work usage and conversation success argues for sticking with a tool long enough to climb its learning curve, rather than tool-hopping.
Related
- 2026 AI-Work Restructuring — the cross-topic synthesis on how AI is reshaping work; this survey’s “most-automated users are the most optimistic” finding is direct first-party evidence for that thread.
- Measuring AI Agent Autonomy in Practice — the companion Anthropic research line on agentic-task autonomy, the same shift (chat → long-running agentic tasks) that motivated the Cadences pipeline change.
- MirrorCode (Epoch AI + METR) — the sibling June-2026 measurement landing the same week: where Cadences measures how people use Claude, MirrorCode measures how far autonomous coding goes (Opus 4.7 leads at a 56% solve rate).
- AI Industry Research — topic hub; sits alongside Stanford HAI AI Index, Gartner, and WalkMe as a recurring first-party/independent measurement series.
- Claude AI — Claude Code and Cowork (the surfaces whose growth reshaped the usage being measured) are catalogued here.
Try It
- Read the cadence chart for your own audience. Before scheduling AI-assisted content or support coverage, sanity-check it against the published rhythms (work tapers on weekends, news in the morning, tax around deadlines) at the report URL.
- Profile your team by automation posture, not just adoption. The survey’s signal is that how automated a user’s workflow is predicts their outlook — segment heavy delegators vs. light users when gauging internal AI sentiment.
- Cross-read with the work-restructuring synthesis. Pair this with 2026 AI-Work Restructuring to connect the survey’s optimism finding to the broader labor-market narrative.
- Pull the full report for the numbers behind the framing. This article is a partial extract; the full report, appendices, and PDF (including Chapter 2) are at anthropic.com/research/economic-index-june-2026-report.
Open Questions
- This is a partial extract. Key framing, Chapter 1, and Chapter 3 are captured here; Chapter 2 is not fully captured. The full report, appendices, and downloadable PDF live at https://www.anthropic.com/research/economic-index-june-2026-report.
- Causation vs. selection in the optimism finding. Does using Claude in the most automated way cause optimism about pay/security/meaning, or do already-optimistic workers self-select into heavy delegation? The report links usage to survey responses but the directionality is not resolved here.
- Are the secondary stats exact? The “nearly half expect major job-responsibility changes in 12 months,” “more concern for junior coworkers,” and “recipes in the evening / gardening stable” figures come from Anthropic’s launch thread (secondary synthesis) and the 6+-month tenure lifts come from a third-party summary — none verified line-by-line against the primary report body.
- How does Cadences compare to prior editions’ findings? Earlier reports (Sept 2025, Nov 2025, Jan 2026, March 2026) tracked automation/augmentation share and “economic primitives” from seven-day samples; whether the daily-sampling method revises any of those trend lines is not addressed in this extract.