Reading the Rhythm of AI Work: What the Anthropic Economic Index Cadences Report Reveals

TRANSMISSION RECEIVED · PLANET ECON-1 · COORDS [-0.45, 0.21]

Anthropic has turned Claude’s usage logs into hourly samples, giving the AI Economic Index its first real heartbeat. The report is called Cadences because the data carries the breathing pattern of a week, a day, even a tax deadline. For anyone building for users, it reads like a timetable of when people are willing to be reached by AI.

What the Cadences report actually measures

The headline methodological move is sampling at the hour level rather than aggregating by week or month. Previous economic index rounds treated AI usage as a smoothed curve; Cadences disaggregates it, exposing the texture of when demand arrives and what shape it takes. The report classifies conversations along two axes: intent (work versus personal) and output type (more than thirty categories, from marketing copy to recipes to backend architecture). Cross-referencing intent, output, and timestamp reveals not just how much AI is used, but the rhythm of the underlying life and work it serves.

The weekday-versus-weekend split is stark. On weekdays roughly 65 percent of conversations are work-related; on weekends personal use rises from about 35 percent to nearly 50 percent. Execution-heavy tasks like backend architecture and API debugging drop fastest on Saturday and Sunday, while AI agent design, quantitative trading, game development, and startup-related conversations climb. In other words, weekends are when users open AI to think about new things rather than to ship existing ones.

Inside a single day, demand peaks are sharp and predictable. News-related requests crest around 7 a.m., business email drafting around 10 to 11 a.m., recipe requests hit 2.3x the average at 6 p.m., and sleep advice concentrates before 5 a.m. These are not curiosities; they are real life rhythms projected onto AI. External events carve even sharper spikes: in the days before the U.S. tax deadline on April 15, tax-related conversations ran eight times the baseline, then collapsed on April 16.

Outputs also reveal where value concentrates. Marketing content and blog posts are among the most work-oriented categories (around 80 and 81 percent work intent), and conversations from higher-wage occupations consume more tokens. A marketing manager earning roughly double an editor’s hourly wage consumes about 2.5x the tokens. The report’s clearest signal: AI compute is not distributed evenly, it follows the density of the task.

The report also distinguishes two modes of delegation: automation and augmentation. Heavy automation users, counterintuitively, report feeling more optimistic about AI and believe their skills are appreciating, not depreciating. Tool choice shapes delegation depth: writing a blog post on Claude Code averages a single prompt, while the same task on Chat averages thirteen turns of back-and-forth. The product surface determines how work gets done.

Why marketers should care

For marketing, the report is less a benchmark and more a metronome. If news consumption peaks at 7 a.m. and recipe demand at 6 p.m., then content published off-peak is the right message at the wrong time. The deadline-driven spike pattern (tax, renewals, back-to-school) translates directly to e-commerce peaks: AI-assisted support and content deployed 24 to 48 hours before a deadline captures the most demand. And the token-density finding tells marketing teams that AI budget should track the value of the task, not the volume of calls.

How to use it

• Treat the content calendar as an AI timetable: news and analysis before 7 a.m., B2B email and product updates at 10 to 11 a.m., lifestyle and food content around 5 to 6 p.m., emotional and wellness content late night to early morning.
• Scale support bots before demand peaks and batch-generate content during troughs; deploy deadline-driven help 24 to 48 hours ahead of tax, renewal, or seasonal events.
• Budget by value density: assign stronger models and longer reasoning to strategy and long-form work, lighter models to repetitive tasks, and measure ROI as business value per thousand tokens rather than call count.

// END OF LOG