Labor Vector
Labor Vector meta-aggregator / Labor performed with AI

Anthropic Economic Index

Labor performed with AIGains and lossesreport seriesactive
Anthropic Economic Index: a report series with data releases: six reports and their datasets, from the operator's own product usage. Operator: Anthropic (its Economics team). Its unit: a sampled Claude conversation (Claude.ai consumer chat and Cowork) or a 1P API prompt-response record, classified by Claude to an O*NET task and to economic primitives; since June 2026 also a survey respondent linked to up to 20 of their own sessions. Whose labor: the tasks people bring to Anthropic's own product Claude, mapped to the O*NET tasks of the occupations that typically perform them; observed on Claude.ai, Cowork, Claude Code and Anthropic's first-party API only.
Visit Anthropic Economic Index ↗The /non rowThe DSL audit
The Anthropic Economic Index provides insights into how AI is being incorporated into real-world tasks across the modern economy.

In its own words: https://huggingface.co/datasets/Anthropic/EconomicIndex/raw/main/README.md, read 2026-10-29

Who supplies, who directs, who gains

Supplies the laborClaude users and 1P API customers bring the tasks; Claude performs them, recorded as the share of occupational tasks 'we already see being done with Claude'
Directs itthe user, recorded as collaboration mode (directive, feedback loop, task iteration, validation, learning) and as the autonomy users delegate to Claude on a 1-5 scale; users choose which tasks to bring
Receives the gains, bears the costsnot recorded as a distribution; the reports estimate aggregate US labor productivity effects and record survey respondents' self-reported productivity gains and their expectations about pay, job security and shared gains

Each as the resource itself records it. Where it records nothing, the census says so.

What it records

IntersectionsLabor performed with AI: assistance, delegation, task fidelity, correction burdens and user labor drag; Distribution of gains and losses: wages, employment, workload, credit, ownership, and who captures productivity gains
Kind of evidenceinteraction dataexposure estimateworker testimonydata access
Unit of recorda sampled Claude conversation (Claude.ai consumer chat and Cowork) or a 1P API prompt-response record, classified by Claude to an O*NET task and to economic primitives; since June 2026 also a survey respondent linked to up to 20 of their own sessions
Whose laborthe tasks people bring to Anthropic's own product Claude, mapped to the O*NET tasks of the occupations that typically perform them; observed on Claude.ai, Cowork, Claude Code and Anthropic's first-party API only
Methodprivacy-preserving classification of sampled transcripts by Claude (another instance of Claude reads the transcripts); task mapping to O*NET; classifiers for five economic primitives validated for directional accuracy; since June 2026 continuous daily sampling and a survey of Claude users linked to their usage op
Coveragegeography: global Claude usage, with country and US-state breakdowns from IP geolocation; period: reports from February 2025 to June 2026; the January 2026 report samples 13-20 November 2025; the June 2026 report samples 10 April to 10 June 2026
What it findsaugmentation 52% and automation 45% of Claude.ai conversations in November 2025; 1P API traffic three-quarters automation; top ten tasks 24% of Claude.ai conversations and 32% of 1P API traffic; the most common task is modifying software to correct errors; 49% of jobs have seen AI usage for at least a quarter of their tasks (across reports); removing the tasks Claude covers produces a net deskilling effect across most occupations; implied labor productivity growth of 1.8 percentage points a year falls to about 1.0-1.2 once task success is factored in; survey: 10% of respondents rate losing their own job as likely or very likely; those who use Claude in the most automated way are the most optimistic about pay and job security
Its stated limitstask success is Claude's own assessment; primitives are directionally accurate and not exact; observed success rates reflect users' selection of tasks; the survey is not representative of the general population; the primitives are not shown to be causal; the implications depend on how often conversations displace or augment work

op the resource's own statement · 3p a third-party account · obs our observation · rd our reading. Hover a tag for its quote; each opens below with its source.

What this evidence can settle

Can answer

  • which O*NET tasks are performed with Claude, and in what share
  • how users collaborate with Claude (automation and augmentation modes, delegated autonomy) by task, occupation, country and product surface
  • how Claude-assessed task success and estimated time savings vary with task complexity
  • what Claude users report and expect about their own work, linked to their usage

Cannot answer

  • AI use on other providers' systems
  • observed employment or wage change
  • whether a conversation displaces or augments work that would otherwise be done by a person
  • who captures the productivity gains it estimates
  • task success as judged by the user or by an external grader

In the worked questions

Where it meets Labor Vector

rd The archive's reading: the collaboration modes and the AI-autonomy and task-success primitives classify who directs a Claude task and whether Claude completed it, as Claude assesses its own completion; this is the nearest of the resource's measures to DSL's question of whether the system's labor advanced the commissioned task. Its turn counts (users engage more, 1.53 times as many turns, in higher-wage occupations) and task iteration mode bear on user labor drag as the resource records them.

Entity

Canonical nameAnthropic Economic Index
Also known asAEI, EconomicIndex (the Hugging Face dataset), Anthropic Economic Index report, Anthropic Economic Index Survey, Anthropic AI Usage Index (AUI)
Addresshttps://www.anthropic.com/economic-index
OperatorAnthropic (its Economics team)
Founded2025-02-10 (the initial data release) op
Kindreport series with data releases: six reports and their datasets, from the operator's own product usage
Lifecycleactive — six reports from the initial release of 2025-02-10 to the sixth of 2026-06-26; the index page last updated 26 June 2026; the June 2026 report changed the pipeline to continuous daily sampling
Count1M Claude.ai conversations and 1M 1P API transcripts (January 2026 report); over 3,000 unique work tasks on Claude.ai; about 9,700 linked survey respondents (June 2026 report) (as of 2026-06-26)
Data accessdata releases on Hugging Face (Anthropic/EconomicIndex), with prompt texts in an online appendix op
Licencedata CC-BY; code MIT (the dataset card's YAML header gives license: mit) op
Versions and datesinitial release: 2025-02-10; third report: 2025-09-15 (arXiv 2511.15080, submitted 2025-11-19); fourth report: 2026-01-15; fifth report: 2026-03-24; sixth report: 2026-06-26; index page updated: 2026-06-26; read: 2026-10-29
Observation2026-10-29: read

Relations

Used by

Neighbours

Captures: the composition layer's account

The /non row

Evidenceevery field's quote, its source and the hash of the saved copy46
Negative tagswhat it is not5
  • NOT a measure of AI use across providers: a platform-specific observation of Anthropic's own product, Claude (Claude.ai, Cowork, Claude Code and the first-party API), published by Anthropic
  • NOT an employment outcome series: it observes usage and survey responses
  • task success is Claude's assessment of whether Claude completes tasks
  • the Economic Index Survey is NOT representative of the general population
  • NOT independent of the provider: the transcripts are classified by another instance of Claude
Sourcesevery source read, with its date and hash6
Not yet checkedwhat remains to read4
  • the index home page's data (the saved page shows only 'Loading data'; it is a JavaScript application)
  • the datasets on Hugging Face (only the README was read)
  • the 1st, 2nd and 5th reports, and the 3rd report beyond its arXiv abstract page (its PDF was saved by the /non composer and not admitted), the appendices, and the 'Labor market impacts' and robot-exposure releases
  • the companion Claude Code report and the October 2026 'The AI expectations gap' article listed as related

Cite this record

Lee Sharks, ed. "Anthropic Economic Index." Labor Vector meta-aggregator, laborvector.org, compiled 2026-10-29. https://www.laborvector.org/aggregators/anthropic-economic-index

Record https://www.laborvector.org/aggregators/anthropic-economic-index · entity https://www.laborvector.org/aggregators/anthropic-economic-index#entity · the census as data

@misc{lv-anthropiceconomicindex2026,
  editor = {Sharks, Lee},
  title = {Anthropic Economic Index},
  howpublished = {Labor Vector meta-aggregator, laborvector.org},
  year = {2026},
  note = {Compiled 2026-10-29; every field sourced and dated},
  url = {https://www.laborvector.org/aggregators/anthropic-economic-index}
}
Editor: Lee Sharks (ORCID 0009-0000-1599-0703). This page is the Labor Vector meta-aggregator's record of the resource above: its reading and fields are ours, and each field carries its source, its basis and the hash of the copy read. The resource's own terms and findings are kept as its own. Compiled 2026-10-29. CC BY 4.0.