The work tasks people use Claude AI for most, according to Anthropic

As AI adoption grows, many are wondering how it will impact the labor market long-term (ChatGPT Deep Research has its own predictions). Anthropic is trying to find out.

On Monday, the company published its first Economic Index, which investigates what kinds of employees are using Anthropic’s Claude chatbot and for what types of tasks. Taking a different approach than many AI and work studies hoping to map the future, Anthropic focused on similarities in work tasks rather than job titles, and on actual chatbot queries instead of survey responses.

Also: Crawl, then walk, before you run with AI agents, experts recommend

“Jobs often share certain tasks and skills in common: For example, visual pattern recognition is a task performed by designers, photographers, security screeners, and radiologists,” the company explains in the report announcement. 

Using its own Clio system for privacy, Anthropic analyzed 1 million anonymized conversations Claude Free and Pro users had with the chatbot. Mapping each to the Occupational Information Network (O*NET), a US Department of Labor database of 20,000 work tasks, Clio identified which task “best represented the role of the AI” in each conversation, Anthropic’s announcement explains. Chats were then grouped into job categories such as arts and media, computer and mathematical, and business and financial.

Software engineering tasks made up the majority of the queries in the dataset — 37.2% of conversations had to do with debugging code, network troubleshooting, and more, which is somewhat to be expected given that Claude positions itself as a coding-first model. The next-largest category of queries had to do with writing and editing at 10.3%, which Anthropic grouped as “arts, design, sports, entertainment, and media” jobs like copywriting.

Also: I pitted Claude 3.5 Sonnet against AI coding tests ChatGPT aced – and it failed creatively

The study observed that these two job categories make up only 3.4% and 1.4% of the US economy, respectively — much less than office admin and sales jobs, for example — but use AI at much higher rates. Jobs in science and education also showed higher rates of AI use relative to their saturation in the economy.

Beyond programming, top tasks within these four categories included producing for entertainment like film and TV, conducting research, and creating educational materials.

Augmenting vs. automating work

The report also found that AI augments human capabilities 57% of the time and automates work — directly performs tasks for people — 43% of the time.

Also: Cerebras CEO on DeepSeek: Every time computing gets cheaper, the market gets bigger

“In just over half of cases, AI was not being used to replace people doing tasks, but instead worked with them, engaging in tasks like validation (e.g., double-checking the user’s work), learning (e.g., helping the user acquire new knowledge and skills), and task iteration (e.g., helping the user brainstorm or otherwise doing repeated, generative tasks),” the report clarifies.

However, Anthropic admits that it can’t be sure users were querying Claude for work purposes in these conversations, just that the queries themselves aligned with occupational tasks. This is especially relevant given that the study doesn’t review data from API, Team, or Enterprise users.

Anthropic also can’t be sure whether users took Claude’s written responses or code snippets as is or edited them outside of the application, which makes the difference between augmentation and automation.

Other findings

Using O*NET’s median salary data, the study found that AI use is more common for tasks in “mid-to-high wage occupations” like data science. People in jobs at the lowest and highest wage bands, like salon workers or doctors, were much less likely to use Claude, often because those roles emphasize manual work.

Also: US sets AI safety aside in favor of ‘AI dominance’

“This likely reflects both the limits of current AI capabilities, as well as practical barriers to using the technology,” Anthropic added. Additionally, the study found that:

• Approximately 4% of jobs used AI for at least 75% of tasks
• Roughly 36% of jobs had some use of AI for at least 25% of their tasks
• People use AI to augment their work (which Anthropic defines as collaborating with and enhancing human capabilities) 57% of the time, compared with using it to automate their work (perform tasks) 43% of the time. 

Takeaways

Anthropic plans to regularly rerun its analysis to see if certain roles are experiencing more automation. “We’ll be able to monitor changes in the depth of AI use within occupations,” the announcement says. “If it remains the case that AI is used only for certain tasks, and only a few jobs use AI for the vast majority of their tasks, the future might be one where most current jobs evolve rather than disappear.”

The report itself doesn’t make any policy recommendations. “Developing policy responses to address the coming transformation in the labor market and its effects on employment and productivity will take a range of perspectives,” Anthropic says. “To that end, we are also inviting economists, policy experts, and other researchers to provide input on the Index.”

For those curious about the data itself (or in search of a grain of salt), Anthropic has open-sourced the conversations for additional research efforts.