Predictions about generative artificial intelligence (AI) often focus on extreme outcomes, from mass turnover to massive productivity gains. The new usage-level data provided by Anthropic gives you a more accurate picture of what’s happening within your actual workflows today.
In its latest economic indicators report, Anthropic analyzes anonymized Claude.ai and first-party API interactions since November 2025, tracking how AI is used at the task level, how autonomous those interactions are, and how performance changes as tasks become more complex.
Rather than relying on surveys or self-reported recruitment, this report examines millions of real-world interactions and categorizes them into over 3,000 different work tasks. It introduces five “economic primitives”: task complexity, user and AI skill, autonomy, success rate, and intended use.
The data shows that while AI is a useful tool, it is not the job replacement some fear, with companies being cautious about implementing it but open to the possibility of saving employees time. This is consistent with data from PYMNTS Intelligence, which found that roughly 7 in 10 workers who use AI at work say their workplace encourages the use of AI, while less than 1 in 10 say their employer actively discourages its use.
A small share of tasks accounts for a large amount of usage
The report found that even though more than 3,000 unique work tasks were identified in the dataset, the top 10 tasks accounted for 24% of all conversations in Claude.ai. The concentration is even higher in enterprise environments, where the top 10 tasks account for 32% of interactions in first-party API usage, up from 28% in the previous reporting period.
Software-related work dominates this clustering. Software changes to fix errors alone account for 6% of Cloud usage, making it the most common task. More broadly, computer and math tasks together make up around a third of all interactions, highlighting how strongly the use of AI is tied to a wide range of technical activities.
Advertisement: SCROLL TO CONTINUE
Beyond software, usage is expanding but remains uneven. Educational tasks, such as coursework assistance and instructional content creation, currently account for approximately 15% of interactions with Claude, an increase since the beginning of this year. Writing, editing, and other arts and media tasks also increased as a percentage of usage, indicating that overall usage is becoming increasingly diverse, although it remains concentrated in a narrow set of tasks.
Collaboration accounts for most AI interactions
A central finding of the report is that most uses of AI still involve close human oversight. In the November 2025 sample, 52% of Claude’s conversations were classified as augmented, meaning the user interacted with the model, guided the prompts, reviewed the output, and made the final decision, the company said. By comparison, 45% of interactions were classified as automated, reflecting more directive, manual requests.
This balance signals a return to collaboration, following a high proportion of automated interactions in previous reporting periods. Automation is still common in enterprise API traffic, with models embedded in scripted workflows, but even that use is focused on a narrow set of tasks rather than a broader end-to-end process.
uneven recruitment
Geographical adoption reflects existing economic and labor patterns. Overall usage is led by a handful of countries, including the United States, India, Japan, the United Kingdom, and South Korea.
Even within the United States, usage varies. The top five states account for approximately 50% of Claude’s total interactions, despite having only 38% of the working-age population. At the same time, growth is accelerating in states with low usage rates. If current trends continue, Anthropic estimates that per capita usage could converge across U.S. states within two to five years. This is a much faster pace than the proliferation of many early commodity technologies.
This finding is consistent with broader employee sentiment. As PYMNTS reported on the OpenAI survey, 75% of workers say AI has improved the speed or quality of their deliverables.
