Study reveals how AI agents are being used

AI agents are not mainly being used to book hotels or handle household chores. A study by Perplexity and Harvard, based on hundreds of millions of anonymized interactions, indicates that 57% of their activity is related to cognitive work: researching, learning, analyzing information, and moving professional tasks forward.

The analysis examined how people use Comet and Comet Assistant, Perplexity's agents, to understand who adopts them, how often they use them, and which tasks they delegate. The result offers a different picture from that of a digital assistant that only saves you minor inconveniences.

More thinking partner than digital butler

Thirty-six percent of the most frequent tasks involve productivity and workflows. Another 21% relate to learning and research. In practice, many people use the agent to gather information, organize it, and prepare an initial analysis before making a decision themselves.

The researchers found examples such as these:

• A procurement professional analyzes case studies to identify relevant use cases before speaking with a supplier.
• A student goes through course materials and asks for help understanding what they are learning.
• A finance worker filters stock options and analyzes investment information.

In these cases, the agent does not make decisions autonomously. It handles the initial search and synthesis, leaving the user better prepared to assess, choose, and act.

That distinction matters. The pattern observed is less about avoiding work than increasing your ability to handle difficult tasks.

Use changes over time

A person's first interaction with an agent is usually simple and low-risk: movie recommendations, general questions, or travel plans. But according to the study, many users eventually bring it into tasks that are more useful in their daily lives.

Someone might start by asking about a tourist destination and, weeks later, use the agent to debug a Python program or summarize a financial report. Productivity and workflow tasks have the highest retention rates. In addition, people who start with learning or research are more likely to become regular users.

The progression recalls that of personal computers. They were initially marketed with games or home applications, but became essential tools through spreadsheets and word processors. Something similar could be happening with AI agents.

Who uses them and who depends on them

Adopting an agent does not necessarily mean integrating it into your routine. The difference lies in usage intensity. The study finds that six major professional groups account for 70% of all activity.

Technology professionals generate the largest volume, at 30% of queries. However, marketing, sales, management, and entrepreneurship professionals show higher usage frequency relative to their level of adoption. Once they incorporate the agent, they tend to use it more consistently.

The context also changes the type of help people seek:

• In finance, 47% of queries are devoted to productivity tasks.
• Among students, 43% focus on learning and research.
• In design, tasks related to media and content predominate.
• In hospitality, travel planning has an especially large share.

More than half of the total volume of queries comes from personal rather than professional contexts. This shows that the same agent can serve as a research engine for a student, a work assistant for an analyst, or a planning tool for someone organizing a trip.

For you, the practical takeaway is clear: an agent's value does not depend only on how well it answers a single question. It depends on whether it can fit into repeated tasks and help you move from information to a decision. What remains to be seen is whether this pattern appears outside Perplexity's products and how the way people study and work changes as these agents become part of more routines.