Perplexity–Harvard study suggests that AI agents are becoming cognitive work partners and not digital concierges

Perplexity and Harvard researchers have published the first large-scale study of how people actually use AI agents in the real world. 

The study, titled "The Adoption and Usage of AI Agents: Early Evidence from Perplexity", penned by Jeremy Yang, Noah Yonack, and others answers three fundamental questions: Who is adopting AI agents? How intensively are they using them? And what tasks are they delegating to their AI assistants?

By analysing hundreds of millions of anonymised interactions from Comet and Comet Assistant users, the researchers have uncovered fundamental patterns about agent adoption and usage, and turns out, the findings challenge some common narratives. 

"For instance, agents are good at booking hotels or handling rote chores, but they’re also serving as partners in deep cognitive work. They are reshaping how we learn, work, and solve problems," the team stated.

The researches debunked the notion of AI agents being “digital concierge”, in which humans offload simple tasks to save time, and instead noted that after classifying millions of user interactions, 57% of all agent activity focuses on cognitive work. "Thirty-six percent of the most common tasks are classified as productivity and workflow tasks, with another 21% classified as learning and research."

The study used the following cases as an example: A procurement professional used the Comet Assistant to scan customer case studies and identify relevant use cases before working with a vendor. A student asked the agent to navigate through course materials and analyse what they were learning. A finance worker delegated the tasks of filtering stock options and analysing investment information. "In each case, the agent handled information gathering and initial synthesis autonomously, giving the user the information they needed to make final judgments and decisions, and implementing those decisions at their request," the study said, suggesting that people are not using Comet Assistant to avoid work, they are using it to do better work.

One of the most revealing patterns in the study is the journey of the user where researchers compared how a user uses an agent on Day 1 versus on Day 100. New users often test the waters with low-stakes queries. "They ask about travel plans, movie recommendations, or general trivia. Gradually, there’s a shift. Our analysis of query transitions shows a strong gravitational pull toward productivity. A user might start by asking about a vacation spot, but once they use the agent to debug a Python script or summarise a financial report, they rarely go back."

The productivity and workflow categories have the highest retention rates, and users who engage in learning or research tasks early on are much more likely to become long-term active users. This apparently mirrors the early days of the personal computer which was often sold as a tool for recipe management or games, but it became indispensable because of spreadsheets and word processing—and AI agents are following the same trajectory.

The researchers also noted the gap between adoption rates and usage intensity, which has revealed the progression of technology shifting from a novelty to a necessity. Six core occupations now drive 70% of all agent activity, with certain sectors like digital technologists naturally leading the volume (30% of queries), knowledge-intensive fields like Marketing, Sales, Management, and Entrepreneurship, demonstrate the highest "stickiness." Their usage intensity outpaces their adoption numbers as they integrate the assistant into their daily workflow.

Users also deploy the Comet Assistant to solve the specific friction points of their industry, while Finance professionals are heavily focused on efficiency, dedicating 47% of queries to productivity tasks. Students are focused on utility, with 43% of tasks allocated to learning and research. In design and context-specific usage dominates.

"In the short term, use cases exhibit strong stickiness, but over time users tend to shift toward more cognitively oriented topics. The diffusion of increasingly capable AI agents carries important implications for researchers, businesses, policymakers, and educators, inviting new lines of inquiry into this rapidly emerging class of AI capabilities," the report said.