This insight summarizes Perplexity's June 2026 research piece "How AI Agents Reshape Knowledge Work", produced with Harvard Business School. The study compares how people use Perplexity's traditional Search product against Comet (its agentic "Computer" product), and finds that agentic AI is changing not just how knowledge workers do tasks, but what tasks they take on.
The shift: from manual execution to supervisory oversight. With Search, users translate AI answers into action across multiple tools themselves. With agents, the AI executes the multi-step workflow autonomously while the user specifies the goal and reviews the output. The data is striking: Computer sessions average 26 minutes of machine execution versus 33 seconds for Search — a 48× increase in autonomous work per session. Only 3.7% of Computer sessions were abandoned by the user; 13% paused for clarification.
Knowledge work expands in both breadth and depth. When agents handle execution, users tackle harder and broader problems. 76% of Computer queries involve higher-order cognition (per Bloom's Taxonomy) versus 55% for Search. Computer tasks span 2.40 knowledge domains on average versus 1.74 for Search — a 38% increase in cross-disciplinary scope. Users also work outside their primary occupation 59% of the time with Computer (versus 50% with Search), suggesting agents lower the activation energy for stepping into adjacent domains.
Economics are dramatic. The research reports an 87% reduction in task time (269 minutes to 36) and a 94% reduction in combined model + labor cost for agent-completed work.
Hospitality implication. Roles defined by tool fluency and narrow specialization compress fastest. A revenue manager, guest-experience lead, and service-recovery coordinator could increasingly be one person directing agents across all three workflows — provided the operator has clean data, well-scoped goals, and the judgment to review agent output critically.