Perplexity AI, the company behind the AI answer engine, has released a large-scale joint study with Harvard researchers that examines how people actually use AI agents in the real world[1]. After analyzing hundreds of millions of anonymized interactions, the study found that 57% of usage falls under "cognitive work," suggesting that AI agents are settling in not as task runners but as thinking partners that support how users learn, work, and make decisions[1][2].

A statistical look at hundreds of millions of Comet and Comet Assistant interactions

The newly released research analyzes hundreds of millions of anonymized interactions from users of Perplexity's AI browser Comet and Comet Assistant[1][2]. The research team focused on three questions: who is adopting AI agents, how deeply they are using them, and what kinds of tasks they delegate to AI assistants[1].

All data analyzed in the study was anonymized, and no personally identifiable information was included, according to Perplexity[1]. The full report has been published on Perplexity's official blog under the title "How People Use AI Agents"[2].

57% of usage is cognitive work, split into 36% productivity and 21% learning

One of the most striking findings is that 57% of all usage falls into the category of cognitive work[1]. The breakdown is as follows[1]:

  • 36% for productivity and workflow improvement
  • 21% for learning and research support

Perplexity positions these as tasks that extend human capability, distinct from "digital concierge" uses such as booking a hotel or handling errands[1]. Concrete examples cited include procurement specialists asking the AI to prepare background research for case analysis, students using the AI to organize key points from lectures, and finance professionals leveraging it to compile and analyze stock information[1].

In each case, the AI agent autonomously handles information gathering and initial summarization, while the human user retains responsibility for the final judgment, Perplexity notes[1].

From entertainment to practical use, echoing the early PC era

The study also identifies clear patterns in how usage evolves over time. According to Perplexity, new users tend to start with low-risk queries such as travel planning or trivia, but once they apply the AI to high-value tasks like code debugging or summarizing business reports, behavioral data shows that productivity and learning use continues to grow afterward[1].

Perplexity likens this trajectory to the way personal computers evolved from a hobbyist's machine to a piece of business infrastructure, arguing that AI agents are following the same path toward becoming an everyday practical tool[1].

Six occupations drive 70% of usage, and patterns vary by industry

On the user side, the study emphasizes that the depth of daily use matters more than raw adoption rates[1]. When measured by how deeply people actually use AI agents, six core occupations account for 70% of total usage, with marketing, sales, management, and entrepreneurs standing out for embedding AI agents into their daily workflows[1].

Usage patterns also differ by industry[1]:

  • Finance: 47% of usage goes to productivity tasks
  • Students: 43% of usage goes to learning and research
  • Designers and hospitality: usage centers on industry-specific tasks

Personal use accounts for more than half of all queries, leading Perplexity to describe Comet Assistant as a general-purpose assistant that takes on different shapes depending on the context[1].

Perplexity frames the shift as a "Hybrid Intelligence Economy"

Perplexity positions the study as the first empirical evidence of a transition toward a "Hybrid Intelligence Economy," in which AI agents extend rather than replace human cognitive labor[1]. The fact that cognitive tasks dominate, the company argues, shows that AI is shifting into a role that supports higher-level judgment and creative work rather than substituting for human business activity[1].

Perplexity was founded in 2022 by engineers previously at OpenAI, Meta, Quora, Bing, and Databricks, and currently answers more than 200 million questions per week[1]. The company's core product is a conversational AI search engine that retrieves answers from real-time, trusted sources and returns responses with citations[1].

Summary

The joint study from Perplexity and Harvard provides hundreds-of-millions-scale data to back up a shift in which AI agents are moving from "digital butler" to "thinking partner." The headline numbers — 57% of usage on cognitive tasks and six occupations accounting for 70% of total usage — suggest that AI agents are entering a phase where they function as a piece of business infrastructure. The next question is how quickly society as a whole can adapt to ways of working and learning that assume an AI agent is always involved.

Source: [1] https://prtimes.jp/main/html/rd/p/000000018.000157647.html

Source: [2] https://www.perplexity.ai/ja/hub/blog/how-people-use-ai-agents