Anthropic published the June 2026 edition of its Economic Index on June 26, offering its most detailed picture yet of how people actually use AI at work. The report, subtitled “Cadences,” combined hourly usage tracking with a direct survey of 9,700 Claude subscribers, linking each person’s responses to their real session history through privacy-preserving analysis — a methodology Anthropic says is a first for this type of AI labor research.
What workers say now
Roughly half of survey respondents said AI can already handle 50 percent or more of their work tasks. Four percent said Claude could perform their entire job today. Looking 12 months out, more than one-third expected AI to take over “most or nearly all” of their tasks.
Those figures carry a built-in limitation: only active Claude users were surveyed. Workers whose roles are being substituted rather than augmented are structurally absent from a sample that requires Claude usage to enter.
The rhythms of AI use
By tracking requests at hourly resolution, Anthropic identified clear daily patterns. Users ask for news most often around 7 a.m., switch to recipe help by 6 p.m., and seek sleep advice around 5 a.m. Weekend usage climbs to just under 50 percent of total volume, compared with roughly 35 percent on weekdays — pointing to growth in personal and household use alongside professional applications.
Seasonal spikes are equally sharp: tax-related queries surged eight times above baseline on April 14, the day before the U.S. tax deadline, then dropped immediately afterward. Academic topics show similar exam-period peaks.
What AI actually produces
Ninety-three percent of Claude conversations result in what Anthropic classifies as an identifiable artifact. Explanations top the list at 17 percent, followed by documents at 15 percent and guidance at 11 percent. Conversations in higher-wage occupations consumed 2.5 times more tokens on average than those in lower-wage roles, suggesting the depth of AI assistance scales with the complexity of the job.
A persistent gender gap
Women made up only 12 percent of the linked survey sample, far below their share of the broader workforce. They also used Claude Code at a rate 7.3 percentage points below men. Anthropic flagged the disparity explicitly, calling it an unresolved gap across the AI ecosystem.