AI Agents Outperform Search Engines in Autonomous Work Duration

A groundbreaking study conducted by Harvard and Perplexity has shed light on the capabilities of AI agents in the realm of knowledge work. By analyzing data from two of Perplexity's products—Search and Computer—the research provides a comparative view of their performance. The study, spanning from February 27 to May 27, 2026, focused on 10,000 matched session pairs where tasks were performed using both products. The findings indicate that the Computer agent, which executes tasks autonomously, completed an average of 26 minutes of work per session, while the Search tool managed only 33 seconds. This stark contrast highlights the efficiency of AI agents in handling complex tasks. Moreover, the study noted that the adoption of Computer led to an increase in Search queries, suggesting that these tools complement rather than replace each other. The research also revealed that while Computer sessions involved more demanding tasks, they maintained a lower dissatisfaction rate among users compared to Search. Overall, the data suggests that AI agents are not only more efficient but also enhance the quality of knowledge work.