New Study from Harvard and Perplexity Reveals AI Agents' Dominant Role in Autonomous Work
In a recent breakthrough, a comprehensive study conducted by Perplexity and Harvard University has unveiled key findings regarding the role of AI agents in today's knowledge economy. The research demonstrates that AI agents perform significantly more autonomous work than traditional search engines, offering valuable insights into their impact on productivity.
Context
The study analyzed data from Perplexity's Search and Computer products over a 90-day period from February 27 through May 27, 2026. The team leveraged production data to compare the performance of these AI agents against each other. They focused on query pairs with high cosine similarity above 0.99 to ensure accurate comparisons.
Technical Details
The core methodology involved matching near-identical query pairs between Search and Computer products, ensuring that every session pair was effectively the same task attempted both ways. This approach allowed researchers to observe real autonomous work performed by agents.
Consequences
Autonomous Work Performed
Computer demonstrated a remarkable autonomy rate of 26 minutes per session for performing machine work alone, compared to Search's mere 33 seconds. These findings highlight the significant difference in efficiency between traditional search engines and AI-driven agents.
Cost-Structure Analysis
The study employed a simple task-based model to evaluate cost structures associated with autonomous work performed by both products. It found that agents create a breakeven step count, where their lower marginal costs outweigh fixed costs, making them more efficient than human agents alone.
Our Take
Efficiency and Time Savings
Computer's efficiency is evident in its ability to complete tasks faster—26 minutes versus 33 seconds for Search. This efficiency translates into substantial cost savings. A matched analysis found that Computer sessions led to a 55% reduction in meaningful dissatisfaction rates, indicating better user satisfaction.
Connector Usage
AI agents also excel in task execution by invoking external tools more frequently than traditional search engines do. This automation of repetitive tasks allows users to focus on more complex and time-sensitive activities.
Conclusion
The findings of this study underscore the transformative potential of AI agents in reshaping knowledge work. By automating routine tasks, these agents not only enhance productivity but also improve user satisfaction by reducing manual effort and improving efficiency. As organizations continue to evolve their digital strategies, embracing AI-driven automation will be crucial for maintaining competitiveness in today's fast-paced business environment.
SOURCE: MarkTechPost | https://www.marktechpost.com/2026/06/08/a-new-study-from-harvard-and-perplexity-finds-ai-agents-perform-26-minutes-of-autonomous-work-per-session-vs-33-seconds-for-search
Source: MarkTechPost. AI-assisted editorial synthesis — TechnoExpress.