Multi-Agent Computer Use
Summary
This paper proposes a multi-agent computer use (MACU) system that uses a manager model to decompose tasks into directed acyclic graphs for parallel execution by subagents. It demonstrates consistent improvements over single-agent baselines on multiple benchmarks and better test-time scaling.
View Cached Full Text
Cached at: 06/02/26, 03:37 PM
Paper page - Multi-Agent Computer Use
Source: https://huggingface.co/papers/2606.01533
Abstract
Multi-agent computer use systems outperform single-agent approaches on complex tasks by enabling parallel execution and dynamic task decomposition through directed acyclic graphs.
Computer use agents(CUAs) today are primarily deployed as single serial agents. This setup is suboptimal for complex long-horizon tasks that benefit fromtask decomposition,parallel execution, and consistent re-planning based on new information. In this paper, we argue that we should instead move towards evaluating and building multi-agent computer use (MACU) systems. These systems, which emphasize planning andparallel execution, alleviate many of the shortcomings of single-agent CUAs. We propose a general multi-agent setup in which a manager model decomposes computer use tasks as adirected acyclic graph(DAG), encoding relevant dependencies and goals for subagents. At each iteration, the manager dispatches parallel CUA subagents to carry out nodes on the ready frontier of the DAG, and continuously revises the DAG (adding, canceling, or rewriting nodes) as new findings arrive from subagents. This design treats the partially observable environment of computer use as a first class challenge: information that downstream agents may not be able to re-observe are retained and passed forward through the manager and DAG structure. We demonstrate that MACU consistently improves over strong single-agent baselines by 3.4-25.5% on desktop (OSWorld) and web navigation (Online-Mind2Web, WebTailBench, Odysseys) benchmarks, exhibits more favorable test-time scaling, and solves complex long-horizon tasks where single-agent CUAs get stuck. On Odysseys, a long-horizon web navigation benchmark, MACU improves average task completion wall-clock time by {sim} 1.5 times, demonstrating its efficacy in speeding up traditionally slow CUA pipelines. Our findings highlight that multi-agent coordinationis a promising axis for scalingcomputer use agentsto work productively for longer and more effectively. We release all code and interactive visualizations at https://jykoh.com/multi-agent-computer-use.
View arXiv pageView PDFProject pageAdd to collection
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2606.01533 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2606.01533 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2606.01533 in a Space README.md to link it from this page.
Collections including this paper1
Similar Articles
Recursive Multi-Agent Systems
This paper introduces RecursiveMAS, a framework that extends recursive scaling principles to multi-agent systems for improved collaborative reasoning efficiency and accuracy. It demonstrates significant speedups and token reduction across various benchmarks compared to standard baselines.
TMAS: Scaling Test-Time Compute via Multi-Agent Synergy
TMAS introduces a multi-agent framework that enhances large language model reasoning by scaling test-time compute through structured collaboration and hierarchical memory systems. The approach uses specialized agents, cross-trajectory information flow, and hybrid reward reinforcement learning to improve iterative scaling and stability on challenging reasoning benchmarks.
Simple Multi-Agent Architecture Running Across Our Entire Org. Keeping everything in Loop.
This article describes a multi-agent architecture running at scale, using LangGraph, CrewAI, and Harbor to handle goal agents, task coordination, and secure access with tracing.
How we built our multi-agent research system
Anthropic details the architecture and engineering principles behind its new multi-agent research system, highlighting how parallel subagents using Claude Opus 4 and Sonnet 4 significantly outperform single-agent approaches in complex research tasks.
Securing Computer-Use Agents: A Unified Architecture-Lifecycle Framework for Deployment-Grounded Reliability
This academic paper proposes a unified architecture-lifecycle framework for securing computer-use agents (CUAs) as they transition from benchmarks to real-world software environments. It analyzes reliability challenges across perception, decision, execution layers and creation, deployment, operation, maintenance stages.