The assumption that artificial intelligence is replacing jobs at a rapid pace is being challenged by new data from Anthropic’s Economic Index.
The report, based on millions of interactions with the Claude AI model, suggests that AI is predominantly being used as a collaborator rather than an autonomous worker. 57% of AI-assisted work falls under augmentation—where AI enhances human productivity—while only 43% of tasks involve full automation.
These findings provide a rare glimpse into how AI is integrated into workplace environments, moving beyond speculation and into real-world impact.
According to Anthropic, “Though this analysis by no means reflects all AI usage, it gives us the clearest picture yet of how AI is being incorporated into real-world tasks across the modern economy.” The dataset used for the study has also been open-sourced, allowing external researchers to analyze and verify the trends outlined in the report.
The study challenges widespread concerns about AI-driven job loss, demonstrating that, at present, AI is largely being integrated as a support tool rather than as a replacement for human labor. This has major implications for industries, policymakers, and businesses looking to better understand AI’s role in the modern economy.
Software Developers and Writers Lead AI Adoption
The Economic Index categorizes AI adoption across various job sectors and found that software engineers are the biggest users of AI. 37.2% of all recorded AI interactions come from programming-related tasks, where developers use AI for debugging, optimizing code, and troubleshooting software issues.
This suggests that AI is being leveraged to enhance efficiency rather than replace programmers outright.
Writers and content creators follow as the second-largest group of AI adopters, making up 10.3% of AI-assisted tasks. These professionals commonly use AI to generate drafts, edit text, and structure content, but retain full control over finalizing their work. This aligns with broader trends in creative industries, where AI is being used as an assistant rather than a complete substitute.
Other fields, such as business operations, legal work, and finance, show moderate AI adoption, but sectors reliant on physical labor—such as construction, healthcare, and agriculture—have virtually no AI presence. The limitations of AI in real-world physical tasks remain a barrier to automation in these industries.
Augmentation vs. Automation: How AI Is Actually Used
Anthropic’s report makes an important distinction between AI augmentation and automation.
Augmentation refers to tasks where AI works alongside humans to enhance productivity, such as generating suggestions, providing research assistance, or debugging code. Automation, on the other hand, describes cases where AI completes a task entirely, such as transcription or formatting documents.
“The distinctions between you fully delegating tasks to a language model—versus, like, batting the ball back and forth—are subtle and emerging right now,” said Jack Clark, Anthropic’s co-founder and head of policy.
This highlights the fluid nature of AI’s integration into work environments, where its role is still evolving depending on the complexity of the task.
Anthropic’s findings contradict common fears that AI is eliminating jobs across the board. Instead, the data suggests that most professionals are using AI to refine and improve their work rather than completely handing tasks over to automation.
AI’s Impact on Wages and Job Sectors
The Economic Index also sheds light on how AI adoption correlates with job wages. Contrary to expectations that AI would first replace lower-wage jobs, the data indicates that AI is most commonly used in mid-to-high-wage professions.
Jobs requiring specialized knowledge and analytical thinking, such as programming and data science, see much higher levels of AI use than those at either end of the wage scale.
Low-wage jobs in retail, hospitality, and food service show little AI adoption, primarily because these roles involve hands-on physical work that AI cannot yet perform effectively.
Meanwhile, high-wage professions in law, medicine, and executive leadership also show relatively low AI adoption, likely due to regulatory barriers and the complex, high-stakes nature of decision-making in these fields.
The report suggests that rather than displacing low-income workers, AI is changing the nature of white-collar work by automating specific tasks while leaving human professionals in control of the final output. This insight is key for policymakers and business leaders trying to anticipate AI’s long-term economic effects.
How Clio Tracks AI Usage Across the Workforce
Unlike traditional workforce studies that rely on surveys or market projections, Anthropic’s Economic Index is built on real-world data, gathered through its internal analysis tool, Clio.
This system categorizes millions of Claude interactions, mapping them to job classifications based on the O*NET Database, a widely recognized occupational framework maintained by the U.S. Department of Labor.
By leveraging Clio, Anthropic ensures that its analysis is not based on assumptions but on real data. The tool filters out non-work-related interactions and categorizes AI use based on professional relevance, ensuring an accurate reflection of how AI is affecting workplace dynamics.
Deep Ganguli, who leads Anthropic’s societal impacts team, emphasized the importance of this data-driven approach. “We’d love more eyes on this problem.” He further added, “We want to figure out how the AI industry should make itself legible to the rest of the world.”
As AI adoption continues to grow, Clio’s ability to track changes over time will provide valuable insights into how different sectors integrate AI into their workflows.
Anthropic’s Transparency Push and Open-Source Data Initiative
One of the defining aspects of Anthropic’s research is its commitment to transparency. While most AI companies keep user interaction data proprietary, Anthropic has open-sourced its dataset, allowing researchers to analyze AI’s real-world effects independently. This move sets a precedent for other AI firms to provide more visibility into how their models are being used.
“We want to figure out how the AI industry should make itself more transparent,” Clark stated. “Some of that comes through statements that companies make, but some of it comes through data.” The decision to release AI usage data publicly aligns with broader industry calls for accountability, particularly as governments worldwide work to establish AI regulations.
The European Union’s AI Act and discussions in the United States about AI-related labor protections have highlighted the need for empirical data to guide policy decisions. By openly sharing its findings, Anthropic provides lawmakers and researchers with a factual foundation for shaping future AI regulations.
How AI in the Workplace Could Evolve
While current data suggests that AI is mostly being used to assist human workers rather than replace them, the future of AI in the workforce remains uncertain. Advances in AI capabilities, particularly in areas like multi-step reasoning and autonomous decision-making, could shift the balance toward more automation over time.
However, other factors—including regulatory oversight, business adoption strategies, and public attitudes toward AI—will influence how quickly this transition happens.
Anthropic’s commitment to updating the Economic Index every six months will help track these changes in real time. As AI adoption patterns evolve, ongoing data collection will be essential in determining whether AI remains a tool for augmentation or becomes a more dominant force in automation.
The report reinforces that AI is actively reshaping work but in ways that differ from the prevailing fears of widespread job displacement. By focusing on augmentation rather than full automation, AI is changing how jobs function rather than eliminating them altogether.
For businesses, this means the most effective AI strategies may not involve replacing employees but rather equipping them with AI-enhanced tools to boost productivity.
