Anthropic’s Economic Index lets you see how organizations and individuals are actually using large-scale language models. This report includes the company’s analysis of 1 million consumer interactions and 1 million enterprise API calls on Claude.ai, all from November 2025. The report notes that its numbers are based on observations, rather than, for example, a sample of corporate decision makers or a general survey.
Predominantly limited use cases
Anthropic’s use of AI tends to be focused on a relatively small number of tasks, with the 10 most frequently performed tasks accounting for nearly a quarter of consumer interactions and nearly a third of enterprise API traffic. As you might expect, the focus is on using Claude to create and modify code.
This concentration of the use of AI as a software development tool has remained largely constant over time, suggesting that the value of the model is primarily based on these types of tasks, and that there are no new uses for Claude for other empirically important purposes. This suggests that widespread, general deployments of AI are less likely to be successful than task-focused deployments for which large-scale language models have proven effective.
Extensions outperform automation
On consumer platforms, collaborative use, where users repeatedly query AI over the course of a virtual conversation, is more common than using AI to generate automated workflows. Enterprise API usage shows the opposite, with businesses looking to save money by automating tasks. However, while Claude is successful with short tasks, the more complex the task (or set of tasks) and the more “think time” required, the lower the quality of the results observed.
This means that automation is most effective for well-defined, routine tasks that are simple, require few logical steps, and provide quick responses to queries. Tasks that are estimated to take a human several hours have a significantly lower completion rate than shorter tasks. Successful long tasks require users to iterate and modify the output.
Increased success rates by allowing users to break down large tasks into manageable steps and configure each individually (either interactively or via API).
The company’s observations show that most of the queries it receives for LLM relate to white-collar roles (although poorer countries tend to use claude more commonly in academic settings than, for example, the US). For example, travel agents may delegate complex planning tasks to LLMs and retain elements of more transactional work, while some roles, such as property managers, see the opposite. Routine administrative tasks can be handled by AI, while tasks that require advanced judgment are handled by human experts.
Reliability jeopardizes productivity gains
The report notes that the claim that AI will increase annual labor productivity by 1.8% (over 10 years) is probably best reduced to 1-1.2%, as extra labor and costs need to be taken into account. While a 1% increase in efficiency over 10 years still makes economic sense, the need for activities such as validation, error handling, and rework reduces success rates, so corporate decision makers must make similar adjustments.
The potential benefits for organizations implementing AI also depend on whether the task given to the LLM is complementary or acts as a replacement. In the latter case, the ability of AI to replace tasks normally performed by humans depends on how complex the task is.
It is worth noting that this report found a near-perfect correlation between the degree of user prompting for LLM and successful outcomes. Therefore, how people use AI will determine what it delivers.
Key points for leaders
AI implementations provide value fastest in specific, well-defined areas. Complementary systems (AI + humans) are better than complete automation of complex tasks. Reliability and the additional work required “around” the AI will reduce the expected productivity gains. Changes in the workforce structure depend on the mix of tasks and their complexity, rather than specific job functions.
(Image source: “Virtual Construction Worker” by antjeverena is licensed under CC BY-NC-SA 2.0.)
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expos in Amsterdam, California, and London. This comprehensive event is part of TechEx and co-located with other major technology events. Click here for more information.
AI News is brought to you by TechForge Media. Learn about other upcoming enterprise technology events and webinars.
