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

2606.07489 ↗ · 2026-06-05 · cs.AI, econ.GN

Jeremy Yang, Kate Zyskowski, Noah Yonack, Jerry Ma

The finding

Headline claim

AI agent orchestration (Perplexity Computer) reduces knowledge-work completion time by 87% and cost by 94% versus Search-aided humans, while expanding the scope of tasks users attempt toward higher-complexity, cross-occupational work.

The test

Matched session pairs (cosine similarity > 0.99 on initial queries) drawn from Perplexity Search and Computer users over a 3-month window (Feb 27 to May 27, 2026). Key samples: 10,000 matched session pairs for autonomy and efficiency analysis; 1,000 matched multi-turn pairs for follow-up query classification; 8,000 users across 8 occupation clusters for horizontal scope; 10,000 queries from 5,000 dual-product users for vertical scope. Time and cost estimates combine a tool-based method (logging execution time), an LLM-based procedure, and user interviews.

Effect size & scope

Computer performs 26 minutes of autonomous work per session versus 33 seconds for Search (48x ratio). Average task completion time falls from 269 to 36 minutes (87% reduction); estimated cost falls 94%. Dissatisfaction rate is 55% lower on Computer (1.3% vs. 2.9% medium-to-high dissatisfaction). Scope: Computer queries are 71% abstract non-routine vs. 53% for Search; require 2.40 O*NET Knowledge domains vs. 1.74 (+38%); engage 2.95 Generalized Work Activities vs. 2.24 (+32%); cross occupational boundaries 9 percentage points more often. 23% of Computer task-statement queries involve a task statement absent from the same users' Search history. Adoption: cumulative queries grew 84x over the 3-month window. Effects hold across all 8 occupation clusters for horizontal scope.

Limitations

Authors acknowledge: (1) data comes entirely from a single vendor (Perplexity), creating potential selection bias in who adopts Computer; (2) time and cost estimates rely partly on LLM-based classification and user self-report, not ground-truth measurement; (3) the study period is only 3 months and Computer was newly released, so early-adopter effects may not generalize; (4) output quality is proxied by next-turn dissatisfaction signals rather than independent expert evaluation; (5) the framework is partial-equilibrium and does not model labor-market or organizational equilibrium effects; (6) the "GenAI wall effect" literature suggests agents may not fully close expertise gaps, which the paper does not directly test.

Counterpoint

A fair skeptic would note: (1) the study is authored partly by Perplexity employees, creating obvious commercial interest; (2) the 87-94% efficiency gain depends heavily on assumptions about how long a human would take using Search alone, estimated via LLM and interviews rather than a randomized control; (3) matched session pairs control for query similarity but not for task difficulty, user motivation, or selection into which product a user chooses for a given task; (4) the 84x adoption growth is from a tiny baseline and says nothing about absolute scale or revenue; (5) dissatisfaction measured via next-turn behavior is a noisy proxy for quality.

Economy implication

If the efficiency and scope findings replicate at scale, they suggest AI agents could substitute for significant shares of knowledge-work hours within 12 to 24 months, particularly for multi-step professional tasks that currently require specialist coordination. The cost-reduction magnitude (94%) would, if broadly achieved, compress per-unit labor costs in affected workflows substantially. However, the evidence base is a single vendor's early-adopter population over 3 months, so broad macroeconomic implications remain speculative. Near-term, the findings strengthen the business case for enterprise agent adoption and raise questions about organizational headcount for roles centered on task orchestration and cross-functional coordination.

Who this matters for

Labor & Knowledge Work: The finding that agents push users into cross-occupational and higher-complexity work implies task recomposition rather than simple displacement: coordination and orchestration roles may erode while verification and extension work grows.

Market & Valuation: 94% cost reduction on matched tasks, if it holds beyond early adopters, implies rapid repricing of knowledge-work services and potential revenue pressure on professional services firms.

Technical Capability: The 48x ratio in machine-work time and the scope-expansion results provide field evidence that agent orchestration qualitatively changes the task frontier, not just execution speed.

Bears on (advisory · papers never write evidence)

[1.3] Agentic reliability (multi-step completion without human correction) is improving materially, not just peak single-shot.

[3.1] Inference cost per unit of capability is falling rapidly.

[3.3] Enterprises achieve measurable ROI and renew / expand, not just pilot.

[7.1] Entry-level hiring in AI-exposed knowledge occupations is contracting relative to less-exposed occupations.

[7.3] AI adoption compresses the apprenticeship ladder: fewer junior roles per senior, rather than fewer seniors.

[7.4] Measured productivity gains from AI in knowledge work are large enough to change staffing plans.

Abstract

Frontier AI systems are bridging the gap between intelligence and utility by shifting from conversational assistants to autonomous agents that execute tasks end to end. Using production data from Perplexity's Search and Computer products, we study this transition by examining how AI agents accelerate and reshape knowledge work. Three key empirical findings emerge. First, using sessions with near-identical initial query pairs as natural experiments for the same underlying task attempted with both products, Computer performs 26 minutes of autonomous work per user session, versus 33 seconds for Search. Computer automates task decomposition and execution that Search users might otherwise manually orchestrate and implement. As a result, Computer shifts follow-up query distribution toward higher-order work such as verification and extension. Autonomy also increases execution quality, with per-query dissatisfaction rates 55% lower on Computer than on Search. Second, due to its autonomy advantage, Computer reduces completion time from 269 to 36 minutes on matched tasks, lowering estimated time and cost by 87% and 94%, respectively, compared to humans equipped with Search alone. Third, Computer changes the scope of work that users attempt: Computer queries more often cross occupational boundaries, require higher-order cognition, draw on broader expertise, take the form of composite tasks that bundle interdependent subtasks into a single query, and unlock work activities that are essentially absent from Search usage among the same users. Together, the evidence indicates that AI agents accelerate workflows, enhance output quality, reduce costs, and expand the breadth and depth of automated work.