Perplexity-Harvard Study: AI agents recompose knowledge work

A Perplexity and Harvard study shows Perplexity Computer automates 48 times more work than Search, cutting task time by 87% as AI agents reshape knowledge work.

In collaboration with researchers from Harvard Business School, Perplexity is publishing its first large-scale study on the actual usage of Perplexity Computer, its agent orchestrator launched in February, compared to Search, its answer engine. The analysis focuses on three areas: autonomy, efficiency, and task scope.

Regarding autonomy, the difference is clear. For comparable tasks (10,000 matched pairs), Computer performs an average of 26 minutes of machine execution per session compared to 33 seconds for Search, representing approximately 48 times more automated work, without more frequent abandonment and by chaining more external tool calls via MCP or API. Quality does not decrease: next-turn dissatisfaction drops to 1.3% compared to 2.9%.

Regarding efficiency, the study estimates, for identical tasks, an average reduction of 87% in time and 94% in cost when humans transition from execution to supervision; user interviews mention a median 25-fold acceleration. Crucially, the scope expands: Computer users work outside their original profession in 59% of cases (compared to 50%) and tackle more complex tasks (76% of higher-order cognition queries, compared to 55%). Where Search explains, Computer produces.

The study remains cautious: early observation window, early adopters already familiar with AI, estimates dependent on assumptions about equivalent human time.