AI Agent Adoption Study

Analysis

Why This Paper Matters

This paper is the first large-scale field study to empirically characterize the adoption, usage intensity, and use cases of general-purpose AI agents operating in open-world web environments. As AI agents become increasingly capable and integrated into daily workflows, understanding who uses them, how intensively, and for what purposes is critical for guiding product development, policy, and education. The study leverages hundreds of millions of anonymized interactions from Comet, an AI-powered browser by Perplexity, providing unprecedented scale and ecological validity.

The findings reveal substantial heterogeneity across user segments, with earlier adopters, users in higher-GDP and higher-education countries, and those in digital or knowledge-intensive sectors (e.g., digital technology, academia, finance, marketing, entrepreneurship) more likely to adopt and actively use the agent. This suggests that AI agents may exacerbate existing digital divides unless targeted interventions are made.

Technical Contributions

• Hierarchical Agentic Taxonomy: Introduces a three-level taxonomy (topic, subtopic, task) to systematically categorize agent use cases, enabling structured analysis of what users do with agents.
• Large-Scale Empirical Analysis: Analyzes hundreds of millions of anonymized user interactions, providing robust statistical power and real-world behavioral data.
• Longitudinal Usage Dynamics: Identifies short-term stickiness in use cases but a long-term shift toward more cognitively oriented topics, revealing how user behavior evolves with agent exposure.
• Segmentation Analysis: Quantifies adoption and usage intensity across demographic and economic segments, highlighting disparities.

Results

• Productivity&Workflow and Learning&Research together account for 57% of all agentic queries.
• The two largest subtopics, Courses and Shopping for Goods, make up 22% of queries.
• The top 10 out of 90 tasks represent 55% of queries, indicating concentration in a few high-frequency use cases.
• Personal use constitutes 55% of queries, professional use 30%, and educational use 16%.
• Adoption and usage are higher among earlier adopters, users in higher-GDP and higher-education countries, and those in digital/knowledge-intensive sectors.

Significance

This study provides foundational empirical evidence for the diffusion of general-purpose AI agents, offering actionable insights for researchers, businesses, policymakers, and educators. The taxonomy and usage patterns can inform product design, targeted outreach, and educational curricula. The observed disparities in adoption suggest a need for equitable access initiatives. The shift toward cognitively oriented topics over time indicates that agents may serve as cognitive augmentation tools, with implications for workforce development and human-AI collaboration.