Summary: Anthropic’s second Economic Index report, published on March 27, 2025, analyzed 1 million anonymized Claude conversations from the 11 days after Claude 3.7 Sonnet launched. It found that augmentation still accounted for 57% of usage, while learning interactions rose from about 23% to 28%. A public 630-cluster dataset also showed varied real uses, though the sample covered only Claude.ai Free and Pro users for 11 days.
What Changed After Claude 3.7 Sonnet Launched?
Anthropic published its second Economic Index report on March 27, 2025. The analysis drew on 1 million anonymized Claude.ai Free and Pro conversations gathered during the 11 days that followed the launch of Claude 3.7 Sonnet. Rather than tracking benchmark scores alone, the report set out to answer a practical question: what do people actually do with a more capable model? It follows the first Economic Index, whose data spanned December 2025 to January 2025, and relies on the same privacy-preserving tool, Clio, to map conversations onto tasks in the U.S. Department of Labor’s O*NET database.
Measured against that earlier sample, the share of usage in coding, education, and science rose modestly. Anthropic notes that the coding increase came as no surprise, since Claude 3.7 Sonnet scores higher on coding benchmarks. Growth in education and science could stem from broader AI diffusion, from coding being applied to novel domains, or from unexpected capability improvements. One caveat matters: the report reports shares of usage, not absolute numbers. A category’s share can drop while its absolute usage grows, simply because other categories grew faster. For product builders that distinction is important — a decline in one area does not mean users abandoned it.
The report also notes that computer and mathematical occupations posted the largest absolute increase (+3%), while education and the sciences logged notable percentage gains. That pattern hints at AI moving beyond pure software development into adjacent fields, which may signal new opportunities for tools that bridge coding with domain-specific workflows.
How People Use Extended Thinking Mode
Claude 3.7 Sonnet shipped an “extended thinking” mode that lets the model reason longer on complex questions. According to the report, this mode is used predominantly for technical and creative problem-solving. Tasks tied to computer and information research scientists lead, with almost 10% of usage involving extended thinking, followed by software developers at around 8%. Digital creative roles also show substantial usage: multimedia artists at ~7% and video game designers at ~6%.
Those figures suggest that users switch extended thinking on themselves when they hit tasks demanding deeper reasoning — algorithm design, complex debugging, or creative simulation, for instance. For product builders this is a clear signal: if your tool can tell when a user is tackling a complex task, you might automatically suggest or enable a similar “deep thinking” mode. The data also implies that users are willing to wait longer for a better answer on hard problems, which could inform UX decisions around latency and progress indicators.
Anthropic released a new dataset that maps each O*NET task to its associated thinking mode fraction, available on Hugging Face. It lets researchers and product teams work out which specific tasks drive extended thinking usage, potentially revealing niche opportunities for specialized AI features.
Augmentation vs. Automation: The Balance Holds
The report separates augmentative uses (learning, iterating on outputs) from automation uses (completing tasks outright, debugging). The overall balance is unchanged: augmentation still accounts for 57% of usage. What did shift is the makeup of augmentation. Learning interactions, in which users ask Claude for information or explanations, climbed from ~23% to ~28% of all usage. That growth points to people treating AI increasingly as an educational tool, not only a task executor.
The augmentation-automation split swings widely from one occupational category to another. Community and Social Service tasks, education and counseling among them, approach 75% augmentation. Production, or computer and mathematical, occupations lean closer to 50-50. Tellingly, no occupational category is automation-dominated. For product strategy this is a key insight: most users still want to collaborate with AI rather than hand off entire tasks.
At the task level, copywriters and editors post the highest amount of “task iteration,” where human and model co-write content together. Translators and interpreters rank among the highest for “directive” behavior, where the model finishes translations with minimal human involvement. These cases show that one model can support different collaboration styles depending on the domain. Product builders should consider interfaces that support both iterative co-creation and more autonomous execution, according to what the user is after.
A Bottom-Up Taxonomy of Real Usage
The report also presents a first-of-its-kind bottom-up taxonomy of Claude.ai usage, built with Clio from the same anonymized conversations. It contains 630 granular clusters, arranged in three levels of hierarchy, with descriptions, prevalence metrics, and automation/augmentation breakdowns. Where top-down approaches map usage onto predefined ONET tasks, this bottom-up approach captures use cases ONET might miss.
Some clusters stand out: helping with water management systems and infrastructure projects, creating physics-based simulations with interactive visualization, assisting with font selection and troubleshooting, creating or improving job application materials, providing guidance on battery technologies and charging systems, and handling time zone issues in code and databases. The range shows how varied real-world applications are, and many of them are not obvious from standard job descriptions.
For product builders this taxonomy is a goldmine. It offers a data-driven picture of what people actually do with AI, which can spark new features or point to underserved niches. The dataset is freely available on Hugging Face, so teams can dig into it directly when shaping their product roadmaps.
Limitations and Practical Takeaways
The report carries several limitations. Its sample covers only Claude.ai Free and Pro conversations across 11 days, so it may not generalize to the entire labor market or to other AI products. Anthropic also cautions that O*NET descriptions may not perfectly represent Claude’s actual usage. The occupation “fine artists, including painters, sculptors, and illustrators,” for example, likely involves more digital art than traditional painting or sculpture on Claude.ai. On top of that, the report reports shares of usage rather than absolute numbers, which can mislead if read carelessly.
For product builders, the key takeaways are:
- Extended thinking is a feature users want for hard problems. Think about adding a similar mode that turns on automatically for complex tasks, or at minimum make it simple to enable.
- Augmentation outweighs automation. Most users want to collaborate with AI, not replace themselves. Build workflows that support iteration, learning, and feedback loops.
- Learning keeps growing. More learning interactions imply that educational features, explanations, and tutorials can add real value to AI tools.
- Put the released datasets to work. The bottom-up taxonomy and thinking-mode data are both public. Study them to find genuine use cases and keep your product aligned with actual demand.
To sum up, Anthropic’s second Economic Index report offers a rare, data-rich look at how people use a frontier AI model. It is not a perfect mirror of the whole economy, but it yields actionable insights for anyone building AI products. Once you know which tasks drive extended thinking, where augmentation or automation dominates, and which real use cases surface bottom-up, you can make better calls on features, UX, and positioning. The datasets are open, so the next step is to dig in and find the patterns that matter for your own product.