Anthropic Economic Index: Cadences in AI Usage and Perceptions

As Claude usage shifts from simple chat to long-running agentic tasks via Claude Code and Cowork, traditional transcripts no longer capture the economic impact. This report exposes a central tension: observed usage data (hourly patterns, artifact types, compute consumption) must be reconciled with users’ perceptions and expectations about AI’s role in their work. The authors argue that measuring both logs and self-reports is necessary to understand how AI diffuses into economic life.

The report introduces a higher-frequency data pipeline and a new classifier for conversation outputs, or artifacts. Key findings show that compute scales with the value of work: conversations mapping to higher-wage occupations consume more tokens, and more valuable artifacts (apps, websites) require significantly more tokens than simple explanations. Additionally, the level of autonomy delegated to Claude is higher on Claude Code than on chat or Cowork, driven more by product design than by model choice. Survey results from 9,700 respondents reveal that people who delegate the most are the most optimistic about pay, job security, and skill value, and that reported exposure to AI exceeds observed exposure across occupations.

For builders, the practical takeaway is that AI usage patterns are deeply embedded in daily rhythms and that the most economically valuable tasks demand more compute and autonomy. The survey challenges the assumption that heavy delegation leads to skill atrophy: heavy delegators report learning at the same rate as others. Crucially, the gap in autonomy between chat and agentic products persists even when controlling for model, suggesting that product interface shapes human-AI collaboration more than raw capability. Understanding these cadences will be essential for designing systems that augment rather than displace human work.