AI-Wiki

Anthropic Economic Index: New building blocks for understanding AI use

6 min read

Anthropic Economic Index — New Building Blocks for Understanding AI Use

TL;DR

The fourth report in Anthropic’s Anthropic Economic Index series. Introduces economic primitives — five measurements derived from privacy-preserving classification of ~1M Claude.ai conversations and ~1M 1P API transcripts (sample period: November 2025, predominantly Claude Sonnet 4.5):

  1. Task complexity — estimated human time without AI; whether multiple tasks are handled in one conversation
  2. Human and AI skill level — years of education needed to understand prompts and responses
  3. Use case — work / education / personal
  4. AI autonomy — degree of delegation, from collaboration to fully directive
  5. Task success — Claude’s own assessment of completion

This page summarizes the summary article dated 15 Jan 2026; the full underlying report is referenced but not separately ingested.

Key claims

Tasks

Speedup scales with task complexity (Claude.ai sample):

  • Tasks requiring high-school education (12 years): ~9× speedup
  • Tasks requiring a college degree (16 years): ~12× speedup
  • API speedups are larger still

Success rate is slightly lower at higher complexity:

  • Tasks needing less than high school: 70% Claude success
  • Tasks needing a college degree: 66% Claude success

“Claude’s impact on task speedup scales more sharply with complexity than complexity correlates with a decrease in success rate.”

“AI’s productivity gains are currently accruing in tasks that require relatively high human capital, which is consistent with the evidence that white collar professionals are more likely to use AI at work.”

Time horizons (METR comparison)

The report situates its findings against METR’s task-horizon benchmark for Claude Sonnet 4.5:

SourceDuration at which 50% success is achieved
METR (fixed-task benchmark)~2 hours
Anthropic 1P API~3.5 hours
Anthropic Claude.ai~19 hours

The Claude.ai number is much higher because of two methodological differences explicitly noted in the report:

  • Selection bias — users bring tasks to Claude that they expect Claude to succeed on.
  • Task decomposition with feedback loops — users break complex tasks into smaller steps, with Claude correcting course mid-conversation.

Country-level use

  • Higher GDP-per-capita countries: Claude is used more for work and personal use.
  • Lower GDP-per-capita countries: more educational coursework use.
  • The report describes this as an “adoption curve”: education-heavy in low-income, diversifying toward work and personal as income rises.
  • Cited as consistent with comparable Microsoft analysis.
  • Anthropic’s Rwanda partnership with ALX is framed as a literacy-to-general-use transition program (year-long Claude Pro grants for graduates).

Occupations — coverage and effective coverage

Pooled task coverage:

  • January 2025 first report: 36% of jobs saw Claude used for ≥25% of their tasks.
  • Pooled across reports through Nov 2025: 49%.

Effective AI coverage (a new occupation-level metric introduced in this report):

  • Defined as the share of a worker’s time-weighted duties that Claude could successfully perform.
  • Outliers where effective coverage exceeds task coverage: data entry keyers, radiologists, medical transcriptionists.
  • Outliers where effective coverage is below task coverage: teachers, software developers.

Task-content / education skew:

  • Claude-covered tasks average 14.4 years of predicted education (≈ U.S. associate’s degree).
  • All tasks economy-wide (employment-weighted) average 13.2 years.

Aggregate productivity

  • Earlier Anthropic estimate (carried over from prior reports): widespread AI adoption could add +1.8 percentage points to U.S. labor productivity growth per year over a decade.
  • Reliability-adjusted estimate (factoring in task-level success rates):
    • +1.2 pp/yr when measured on Claude.ai
    • +1.0 pp/yr when measured on the 1P API (more challenging task mix)
  • Even +1 pp/yr would return U.S. productivity growth to late-1990s / early-2000s rates.

Updates on previous measures

Task concentration is rising:

  • Top 10 work tasks on Claude.ai now account for 24% of all conversations, up from 21% in January 2025.
  • Computer and mathematical tasks ≈ ⅓ of Claude.ai conversations and ≈ ½ of API traffic.

Augmentation / automation split on Claude.ai over time:

Sample periodAugmentationAutomation
January 202555%41%
March 202555%42%
August 202547%49%
November 202552%45%

The August 2025 sample was the only one where automation led; the November 2025 sample reverses to augmentation-led.

Geography: top countries (US, India, Japan, UK, South Korea) still lead overall Claude.ai use, with adoption well-explained by GDP per capita. Within the US, distribution across states has become noticeably more even — the report’s model implies state-level equalization within 2–5 years if the trend holds.

Methodology notes

  • Privacy-preserving classification — Claude itself answers a fixed set of questions about each conversation in the sample. Full methodology in chapter 2 of the underlying report (not separately ingested here).
  • Sample size — 1M Claude.ai Free / Pro / Max conversations + 1M 1P API transcripts.
  • Sample period — November 2025.
  • Predominant model — Claude Sonnet 4.5. The report acknowledges the subsequent release of Claude Opus 4.5 may shift future numbers.
  • Task taxonomy — O*NET.

Limitations / caveats (per report)

  • Coverage analysis is restricted to tasks performed on Claude.ai — does not capture other LLM platforms, internal tooling, or non-LLM AI.
  • The deskilling analysis is a first-order effect; the report explicitly does not predict labor-market dynamics will let it materialize unchecked.
  • The Claude.ai task-horizon number embeds selection bias and task decomposition.
  • The report is observational, not causal.

Deskilling — task-composition shift

A specific finding: removing all Claude-covered tasks from current jobs would, as a first-order effect, deskill jobs on average — because Claude-covered tasks skew toward higher-education content. Most-affected professions named: technical writers, travel agents, teachers. Real-estate managers move in the opposite direction. The report frames the analysis as a useful first-order signal, not a labor-market forecast.

“We’re not necessarily predicting that this deskilling will occur: it’s possible that even if AI fully automated the tasks it currently supports, the labor market would dynamically adjust in ways that this analysis doesn’t account for.”

See ai-deskilling for the wiki concept page.

Quotes worth saving

“AI’s productivity gains are currently accruing in tasks that require relatively high human capital, which is consistent with the evidence that white collar professionals are more likely to use AI at work.”

“Claude’s impact on task speedup scales more sharply with complexity than complexity correlates with a decrease in success rate.”

“Our analysis shows how Claude’s effective time horizons might look different to those found in a study with a consistent set of tasks.”

“We expect [tasks] might move from Claude.ai to the API (that is, from predominantly consumers to predominantly businesses) as they become more reliable — and if this happens, it’ll give us another possible indication of coming economic impacts.”

Graph View

Backlinks