Anthropic Economic Index report: Cadences
Understand how Claude use varies by time and context, what people produce with it, how compute relates to work value, and what users think.
An essay summarizing the Anthropic Economic Index report "Cadences," by Massenkoff, Lyubich, Sacher, Hitzig, Zhang, Heller, and McCrory, published June 26, 2026. The shift that forced a new lens A year ago, studying how people used Claude was relatively simple. Almost everything happened in a chat window, one message at a time, and reading a transcript told you most of what you needed to know. That is no longer true. With the growth of Claude Code and Cowork, a lot of usage is now long-running agentic work where Claude operates on its own for stretches at a time. A transcript no longer captures the whole story. So Anthropic rebuilt the measurement. This report introduces three changes. It samples usage far more frequently, down to the hour rather than a single week at a time. It adds a classifier that labels the concrete output of each conversation, which the report calls an artifact. And it breaks results out by surface, separating chat and Cowork (together called Claude conversations) from developer traffic through the direct API. On top of all that, it adds something usage logs can never show, which is what people actually think. A new Anthropic Economic Index Survey, launched in April 2026, asks users directly about their work and their expectations. The report is organized in three moves. First, the rhythms of when people use Claude. Second, what they walk away with. Third, how they feel about all of it. Chapter 1: usage breathes with daily life The headline of the first chapter is almost poetic. Claude usage mirrors the rhythms of the world around it. The clearest signal is the workweek. The share of chat and Cowork conversations that are personal rather than work-related sits around 35% on weekdays and jumps to just under 50% on weekends. The content shifts too. Weekdays lean toward business correspondence, marketing copy, and slide decks. Weekends swing toward emotional support, medical questions, and investment advice. That swing is biggest in high-income countries. The same weekday-to-weekend pattern shows up in Claude Code and direct API traffic, just with lower baseline rates of personal use. Looking inside Claude Code on weekends is revealing. The work that drops the most is the routine engineering, things like backend architecture, API debugging, and data storage. The work that rises the most is the playful and the speculative, like AI agent design, quant trading, and gaming. Weekends also seem to be when people dream. Conversations about starting a business peak on Saturday and Sunday, even as job application activity falls along with everything else work-related. Zoom in further, to the hour, and the day has a texture you would recognize from your own life. People ask for the news around 7 in the morning. Business email tracks the arc of the workday with a slight late-morning peak. Recipe requests spike hard around 6 in the evening, running more than twice their daily average. Media recommendations cluster at night. Sleep advice rises in the small hours just before dawn. There is a sharper economic note inside this. When people do turn to Claude for work at night and on weekends, the tasks skew toward higher-paid occupations. Work mapped to the top two wage quartiles rises during off-hours while the bottom two fall, and that holds even after stripping out computer and math jobs. The likely reason is mundane, which is that marketing managers and programmers are simply more apt to work odd hours. Finally, the calendar leaves fingerprints. The sample period covered US tax season, and the day before the filing deadline, tax-related conversations ran about eight times their normal level, then collapsed the moment the deadline passed. Chapter 2: what people actually take away If chapter one is about when, chapter two is about what. Anthropic classified each conversation by its primary output, sorting more than thirty kinds of artifacts. Almost everything produces something. 93% of Claude conversations yielded a concrete artifact. The most common are explanations at 17%, documents and reports at 15%, and guidance at 11%. Grouped more broadly, conversational outputs like explanations and written deliverables like documents each account for roughly a third of conversations, and code and technical work for about a sixth. But knowing what an output is does not tell you what it is for. The same artifact can be a work deliverable or a weekend project, so the report splits each category into work, personal, and coursework. Some things are almost always personal. More than 80% of creative writing, guidance, and recipes are personal, and the personal creative writing is mostly fanfiction, worldbuilding, and poetry. The small work-related slice of creative writing is more like video scripts and speeches. On the other side, the most work-dominated outputs are marketing content, blog posts, and database queries, each around 80% work. Some artifacts split almost evenly, like plans and translations, which land near half and half. Then comes the most economically interesting finding. Compute tracks the value of the work. Measuring each conversation in tokens, the report finds that conversations mapped to higher-paid occupations consume more compute. Marketing managers earn about twice what editors do and their conversations burn roughly two and a half times the tokens. The relationship is noisy and has odd outliers, pharmacists being a notable one, but the gradient is real. Building an app eats more than three times the tokens of a typical conversation, while a plain explanation uses about a fifth. Crucially, more output from Claude does not mean less from the person. In higher-wage conversations Claude produces more per turn, but users also take more turns and switch on extended thinking more often. The two rise together, which makes the pattern look more like augmentation than replacement. The chapter also measures how much Claude is left to decide on its own, on a scale from none to extreme. Low-autonomy work is the easily specified stuff like math, translation, and simple questions. High-autonomy work is the open-ended stuff like apps, games, and presentations, the kind of judgment-heavy work that has always been hard to automate. The striking result is that the same task gets handed over with much more autonomy in Claude Code than in chat or Cowork. A blog post on chat takes a median of thirteen rounds of back-and-forth. The same blog post in Claude Code often takes a single prompt. This is not just about which model is running, since the gap survives when you compare like model to like model. The product shapes the behavior. One last detail with a nice ring to it. Claude tends to answer slightly above the level it was asked. Output sits about a year of education above the prompt on average, with the widest gap when people describe something to be built and hand the rest to Claude. Chapter 3: what people think is happening to them The third chapter steps outside the logs. It draws on a survey of about 9,700 users, linked to their usage through privacy-preserving methods. It is not a representative sample of the public, skewing heavily toward computer, math, and management roles, and that caveat matters for reading everything that follows. The first finding is a kind of consensus about the future. Almost everyone expects significant AI progress over the next year, and that expectation is remarkably uniform across very different people. The report calls it a rising tide. Close to six in ten respondents think AI will be able to do a larger share of their work next year than today, and over a third expect it to handle most or nearly all of their tasks within a year. Where people diverge is on what that means for them personally. Early-career workers say AI can already do the largest share of their work, and they carry the most worry about losing it. The most experienced workers report the opposite, putting the share of their work AI can do roughly ten points lower than beginners do. When asked what AI will never manage, the seasoned respondents pointed to judgment, context, and the relational parts of their jobs like building trust and managing people. There is a geographic version of the same pattern. People in higher-income countries report that AI can do less of their work, and lower-income economies tend to use Claude in more automated, hand-it-the-whole-task ways. Then the report lands its most counterintuitive result. You might expect the people who delegate the most to AI to be the most afraid of it. The opposite is true. Across all six dimensions the survey asked about, covering pay, job security, finding a job, meaning, autonomy, and human interaction, the heaviest delegators were the most optimistic. The effect was largest for pay and for the ability to find a job. The people handing over the most also felt their skills were becoming more valuable, not less, and reported learning at the same rate as everyone else. The report is careful here, noting these are self-assessments and cannot rule out quiet skill erosion, but it finds no sign of the feared offloading effect. The fears that do exist are mostly fears for other people. About 10% rated their own job loss as likely in the next year, but they were consistently more worried about others, especially junior colleagues, with over a third putting a junior person's odds of job loss above 60%. Productivity gains were close to universal, with large majorities reporting faster, broader, and better work. A sharp gender difference runs through the data. Women made up only 12% of the linked sample and used Claude noticeably differently, with lower automation and Claude Code shares but more active, iterative, collaborative time in chat, even after controlling for occupation. The survey ends on hope. Asked to dream about an AI-shaped economy ten years out, the most common wish was not replacement but collaboration, work that still matters done alongside AI. Just behind it came the wish to automate the drudgery and win back free time, followed by a hope that the gains would be shared widely. The bottom line Strip the report down and three ideas carry it. AI usage has become woven into the actual rhythm of human life, rising and falling with the workweek, the clock, and the calendar. The work people hand to AI is getting more valuable and more autonomous, especially through agentic tools like Claude Code, and the compute spent tracks the value at stake. And the people closest to that frontier, the ones delegating the most, are not the most frightened but the most hopeful, even as they worry on behalf of the people coming up behind them. The throughline is that AI is no longer a thing people visit in a chat window. It is starting to keep the same hours we do, and the more honest question is no longer whether it will do our work, but how we want to share the work that remains.