Artificial intelligence is no longer just a futuristic concept powered by hype, it is reshaping day-to-day life in measurable and profound ways. A new large-scale study on AI Agent Usage, published by Perplexity and Harvard researchers, provides unprecedented insight into how people rely on intelligent assistants to think, learn, and work.
Using anonymized interactions from hundreds of millions of Comet and Comet Assistant sessions, this research maps the real behavior of users in 2025, revealing patterns that go far beyond expectations.
The first finding is clear: AI agents are becoming thinking partners, not just digital butlers. Although popular imagination paints AI assistants as tools for booking hotels, managing chores, or organizing errands, the data contradicts this assumption.
According to the study, 57% of all agent activity falls under cognitive work. Productivity and workflow tasks account for 36% of total usage, while learning and research represent another 21%. This shift demonstrates that people are using AI agents to scale their abilities rather than avoid responsibility.
Examples from real users reinforce this trend. Procurement teams use agents to sift through case studies and highlight relevant insights. Students rely on agents to decode coursework and analyze new concepts. Finance professionals offload stock filtering and data synthesis.
In every scenario, the AI agent handles the groundwork gathering, organizing, and summarizing information, while humans focus on making decisions and executing strategy. This demonstrates a new era of augmented cognition in which AI expands intellectual capability.
Another powerful insight from the study is how user behavior evolves over time. Most people begin with lightweight queries, movie suggestions, travel tips, general trivia. However, as familiarity grows, their queries also mature.
Once users experience the productivity boost of delegating complex tasks such as debugging code or summarizing financial reports, they rarely return to low-stakes usage. Productivity tasks show the strongest retention rates, while users who explore learning and research early on are significantly more likely to become long-term, high-intensity users.
This mirrors the adoption curve of early personal computers. Initially marketed for household management and entertainment, PCs ultimately became essential due to productivity tools like spreadsheets and word processors. AI agents are following that same trajectory, moving from novelty to necessity as their utility becomes undeniable.
The study also dives into who is adopting AI agents and how intensely they rely on them. While adoption rates matter, intensity of usage paints the clearer picture of true value. Six key occupations now drive 70% of all agent activity, with digital technologists leading in volume.
Yet, the strongest stickiness is found in knowledge-centric roles such as Marketing, Sales, Management, and Entrepreneurship. Once these professionals integrate AI agents into their daily routines, their dependence grows rapidly, confirming that agents fill critical gaps in modern workflow.
Context significantly influences AI Agent Usage. Finance professionals concentrate nearly half their queries on productivity. Students dedicate most interactions to learning and research. Designers rely heavily on creativity-related tasks, while hospitality workers frequently request travel-related recommendations.
This context-specific adoption shows that AI agents adapt to user needs, transforming from research engines in academic environments to multi-purpose assistants in professional settings. Notably, personal contexts still dominate more than half of all queries, underscoring the broad everyday applicability of AI assistance.
This research marks an important turning point in understanding how modern knowledge work is evolving. The study suggests we are moving toward a hybrid intelligence economy, a system where human expertise and AI reasoning amplify one another.
The rise of cognitive task dominance is especially telling. As agents improve, productivity-driven usage is expected to deepen, expanding the scope of what individuals and organizations can accomplish.
As 2025 closes, the question is no longer whether people will use AI agents. They already doacross industries, demographics, and personal needs. The real question is how quickly institutions, workplaces, and global markets will adapt to a world where intelligent collaborators play a central role in thinking, learning, and producing. For more cutting-edge insights on AI and emerging technologies, visit ainewstoday.org and stay ahead of the future.
