How People Really Use AI Agents: Insights from Perplexity and Harvard Research

Recently, I came across a comprehensive study by Perplexity and Harvard analyzing hundreds of millions of interactions with the Comet Assistant. The findings resonated deeply with my experience working on business process automation and AI integration for SMEs across Norway and the EU. What stood out most was that 57% of all AI agent requests involve complex intellectual work rather than simple tasks. Within that, 36% focus on productivity and workflows, and 21% on learning and research.

This matches what I see in practice: procurement specialists using AI to analyze client cases ahead of meetings, students breaking down study materials, and financial analysts filtering stocks and evaluating investments. In these scenarios, the AI gathers data and performs initial analysis, while humans make the final decisions. It’s a clear example of hybrid intelligence in action.

The usage pattern also caught my attention. New users often start with casual questions about travel or movies but quickly transition to work-related queries. Those who explore code debugging or report analysis tend to stay within those domains. Productivity-related tasks show the highest user retention, which aligns with my work automating workflows via tools like n8n, Zapier, and Make.

Breaking down by profession: six main specialties account for 70% of activity. Finance professionals dedicate 47% of their queries to productivity; students spend 43% on learning. Marketing, sales, management, and entrepreneurship show the most intense usage after initial adoption.

From a practical standpoint, this confirms that people don’t use AI agents to avoid work but to tackle more complex tasks efficiently. As someone who builds system solutions and integrates AI workflows via API, this signals the growing importance of hybrid intelligence economies where AI scales human cognitive capabilities.

How would I approach this practically? First, collecting and normalizing interaction data is crucial to understand real user needs. Then, integrating AI agents through APIs into existing workflows allows seamless automation. I’d implement automatic scenarios for repetitive analysis tasks and set up monitoring dashboards to track key metrics like usage patterns and retention. Finally, iterating based on feedback ensures continuous improvement and alignment with user goals.

Practical takeaways:

1. Focus AI agent capabilities on supporting complex intellectual work, not just simple queries.
2. Design onboarding flows that quickly guide users from casual to productive tasks.
3. Prioritize API integration for smooth embedding of AI into workflows.
4. Use monitoring to identify high-retention use cases and expand them.
5. Treat AI as a cognitive amplifier, enabling humans to make better decisions, not just automate low-value tasks.

Source: Research by Perplexity and Harvard University on Comet Assistant interactions.