Which Economic Tasks are Performed with AI Evidence from Millions of Claude Conversations

Core Insights - The article presents a novel framework for measuring AI usage patterns across various economic tasks, revealing that AI is predominantly utilized in software development and writing tasks, which together account for nearly half of all AI usage [2][54] - Approximately 36% of occupations employ AI for at least a quarter of their tasks, indicating a broad integration of AI across the economy [2][54] - The analysis shows that 57% of AI usage is focused on augmenting human capabilities, while 43% is aimed at automation, highlighting the dual role of AI in enhancing productivity [2][54] Group 1: AI Usage Patterns - AI usage is concentrated in software engineering roles and writing professions, with these categories showing the highest levels of interaction with AI systems [6][19] - Only about 4% of occupations exhibit AI usage for at least 75% of their tasks, while approximately 36% of occupations show usage in at least 25% of their tasks, suggesting selective integration rather than comprehensive automation [21][54] - Cognitive skills such as Reading Comprehension, Writing, and Critical Thinking are most prevalent in AI interactions, while physical skills show minimal presence [24][28] Group 2: Economic Implications - AI usage peaks in occupations within the upper wage quartile, particularly in the software industry, while both low-wage and high-wage occupations demonstrate lower usage rates [26][28] - The analysis indicates that occupations requiring considerable preparation, such as those needing a bachelor's degree, show the highest representation in AI usage data [27][28] - The findings suggest that AI is primarily used for specific tasks within occupations rather than fully automating entire job roles, indicating an evolution of job functions rather than outright job displacement [50][51] Group 3: Collaboration Patterns - The study categorizes AI interactions into automation and augmentation, with a slight majority (57%) of conversations reflecting augmentative behaviors, where AI enhances human capabilities [30][34] - Automative behaviors are prevalent in tasks such as content generation and coding, while augmentative behaviors are more common in educational and professional communication tasks [31][35] - The distinction between automation and augmentation is crucial for understanding the potential impacts of AI on productivity and worker engagement [51][52]