A report by Anthropic looks at how people and companies used the AI assistant Claude, using a large sample of anonymised conversations.
Despite widespread fears about artificial intelligence (AI) replacing jobs, the technology is currently assisting workers instead of t killing jobs, a study by AI firm Anthropic suggests.
Despite the job worries, the research is more complex as AI was found to reshape jobs differently, depending on the role. It marks a contrast to CEO Dario Amodei’s past comments that AI could wipe out half of all entry-level white-collar jobs.
Instead of focusing only on how often AI is used, the report looked more closely at what kind of tasks it is given and whether it succeeds, introducing a set of measures the authors call “economic primitives”.
These measures examine the types of tasks people use AI for, how difficult those tasks are, the level of education required to understand both the user’s request and the AI’s response, how much autonomy is given to the AI, and how reliably it completes the task.
The authors said these markers are intended to give “a new window for understanding AI’s impact on the economy”.
The study found that 49 percent of jobs can now use AI in at least a quarter of the tasks involved, which is a 13 per cent increase from early 2025.
The report analysed how people and businesses used Anthropic’s AI assistant Claude in November 2025, drawing on an anonymised sample of two million real conversations from its free and paid services.
Usage remains uneven across jobs and economies
According to the report, Claude usage is concentrated among certain tasks, most of which relate to coding.
Overall, AI is most often used for assignments that require higher levels of education than the average in the economy, such as Software development requests, the report found.
Not all office jobs are affected in the same way by AI and the technology can both upskill or deskill workers, it added.
“For some occupations ,it (AI) removes the most skill-intensive tasks, for others the least,” the report stated.
The report also highlights geographic differences. Countries with higher incomes tend to use AI more frequently, and more often for work and personal tasks. Lower-income countries show a higher share of educational use.
While work-related tasks account for the largest share of Claude usage, educational use is highest in countries with lower Gross Domestic Product (GDP) per capita. Wealthier countries, by contrast, show higher levels of personal use.
Anthropic said this reflects different stages of adoption, with users in lower-income countries more likely to use AI for education, while usage in richer economies broadens to include everyday and personal tasks.
Understanding how AI is used is ‘important’
Researchers also examined whether people were using Claude to fully automate a task for them or to "augment" their work.
Automation typically involves giving the AI a task to complete with little interaction, such as translating text into another language. Augmentation involves collaboration, such as drafting and revising a document together.
On the Claude site, 52 percent of work-related conversations involved augmented tasks. That share is down five percent from January of last year.
The report also found that more complex tasks tend to be less reliable. As tasks become longer or more difficult, Claude’s success rate falls, which reduces the amount of time humans ultimately save.
Earlier estimates assumed AI tasks were successful whenever the technology was used. By taking account of errors and the need for human checking and correction, the report reached more cautious conclusions about productivity gains.
“Claude struggles on more complex tasks: As the time it would take a human to do the task increases, Claude’s success rate falls,” the authors noted in the report.
This is the fourth edition of Anthropic’s economic index, which tracks how AI is being integrated into work and its potential effects on jobs and productivity.
Authors of the report argue that understanding how AI is used is as important as measuring how widely it is adopted.
“How willing users are to experiment with AI, and whether policymakers create a regulatory context that advances both safety and innovation, will shape how AI transforms economies,” the report said.
