AI agents, generative AI systems that act autonomously to perform tasks, have repeatedly been promoted by tech companies as a ‘digital concierge’ capable of making purchases, booking travel, or scheduling meetings on behalf of the user, without requiring their constant input.
But new data suggests that this vision is in contrast to how people are actually using AI agents.
While these agents are good at booking hotels or handling rote chores, over 57 per cent of all agent activity in Perplexity’s Comet browser was focused on cognitive work, according to a new study released by Harvard University researchers in partnership with the AI search startup on Tuesday, December 9.
The study, which researchers claimed was the first large-scale effort to understand how people are using AI agents, found that 36 per cent of the most common actions carried out by Perplexity’s AI browser agent were productivity and workflow tasks, while 21 per cent of the tasks had to do with learning and research.
These findings are based on the researchers’ analysis of hundreds of millions of anonymised interactions from Comet and Comet Assistant users. It was focused on three fundamental questions: Who is adopting AI agents? How intensively are they using them? And what tasks are they delegating to their AI assistants?
Notably, productivity and workflow-related tasks also had the highest retention rates. By using AI agents for learning or research tasks early on, people are also much more likely to become long-term active users, as per the study.
These findings come at a time when companies across industries are looking to adopt AI agents. They challenge some common narratives that could potentially reshape how agents could be used to learn, work, and solve problems.
Story continues below this ad
“This study provides the first empirical proof that we are moving toward a hybrid intelligence economy. The dominance of cognitive tasks in our data suggests that AI agents are scaling human cognitive work. As these tools mature, we expect the “stickiness” of productivity tasks to deepen,” Perplexity said in a blog post.
While 2025 was supposed to be the year AI agents took off, the surge in usage or adoption rates has not yet arrived. Several experts such as Andrej Karpathy, a famed AI researcher, have also questioned the progress of agentic AI.
“I’m not actually sure who said this but they were alluding to this being the year of agents with respect to LLMs and how they were going to evolve. I was triggered by that because there’s some over-prediction going on in the industry. In my mind, this is more accurately described as the decade of agents,” Karpathy said on a recent podcast hosted by Dwarkesh Patel.
Who is using AI agents? And how?
Six core occupations drove 70 per cent of all agent activity in the Comet browser, according to the study. While digital technologists led on volume ( accounting for 30 per cent of queries), other fields like Marketing, Sales, Management, and Entrepreneurship, demonstrated the highest “stickiness”, as per the study.
Story continues below this ad
Researchers also found that users are deploying Comet Assistant, Perplexity’s AI browser agent, to solve the specific friction points of their industry. For instance, over 47 per cent of queries from finance professionals were related to productivity tasks. Queries from students, on the other hand, were made up of 43 per cent of learning and research tasks.
The study further highlighted a key trend in user behaviour, where people may start off using the AI agent for low-stakes tasks like movie recommendations or general trivia and gradually shift to more intensive productivity tasks. “A user might start by asking about a vacation spot, but once they use the agent to debug a Python script or summarize a financial report, they rarely go back,” Perplexity said.