How AI Agents Reshape Knowledge Work: Autonomy, Efficiency, and Scope

· Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, AI Economic & Workforce Impact · Depth: Expert, extended

Summary

Perplexity's Computer product, an AI agent orchestration system, significantly reshapes knowledge work compared to its conversational assistant, Perplexity Search. Analyzing production data from February to May 2026, the study found Computer performs 26 minutes of autonomous work per user session, a 48x increase over Search's 33 seconds, leading to 55% lower user dissatisfaction. This autonomy reduces task completion time by 87% (from 269 to 36 minutes) and cost by 94% compared to human-Search workflows. Furthermore, Computer expands the scope of work: queries more frequently cross occupational boundaries (59% vs. 50% for Search), demand higher-order cognition (76% vs. 55%), require broader expertise (2.40 vs. 1.74 O*NET Knowledge domains), and bundle more subtasks. Notably, 23% of Computer queries involve tasks essentially absent from Search usage, indicating new work possibilities.

Key takeaway

For Directors of AI/ML evaluating agentic AI adoption, recognize that autonomous agents like Perplexity Computer fundamentally alter workflow economics. Your teams can achieve 87% time and 94% cost reductions on complex tasks, while empowering workers to tackle higher-order, cross-functional projects previously deemed too costly or specialized. Prioritize agent solutions for multi-step, generative work to maximize efficiency and expand your organization's operational scope.

Key insights

AI agents, by automating multi-step execution, dramatically boost efficiency and expand the scope of complex, cross-domain knowledge work.

Principles

Method

The study used a matched-pair design comparing Computer and Search sessions with near-identical initial queries from dual-product users, augmented by LLM-based classification and user interviews.

In practice

Topics

Best for: CTO, VP of Engineering/Data, Executive, AI Scientist, Director of AI/ML, Consultant

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.