Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CoAct-1: Computer-using Agents with Coding as Actions

Linxin Song, Yutong Dai, Viraj Prabhu, Jieyu Zhang, Taiwei Shi, Li Li, Junnan Li, Silvio Savarese, Zeyuan Chen, Jieyu Zhao, Ran Xu, Caiming Xiong

TL;DR

GUI-based agents struggle with long-horizon tasks due to brittle perception and inefficient action sequences. CoAct-1 introduces a hybrid multi-agent system that couples GUI manipulation with direct programmatic execution via a Programmer and an Orchestrator that dynamically assigns subtasks. It achieves a new OSWorld state-of-the-art 60.76% success rate (100-step budget) and an average of 10.15 steps per task, outperforming prior GUI-centric methods. The results show that incorporating coding as an action enhances robustness, efficiency, and scalability for general computer automation.

Abstract

Autonomous agents that operate computers via Graphical User Interfaces (GUIs) often struggle with efficiency and reliability on complex, long-horizon tasks. While augmenting these agents with planners can improve task decomposition, they remain constrained by the inherent limitations of performing all actions through GUI manipulation, leading to brittleness and inefficiency. In this work, we introduce a more robust and flexible paradigm: enabling agents to use coding as a enhanced action. We present CoAct-1, a novel multi-agent system that synergistically combines GUI-based control with direct programmatic execution. CoAct-1 features an Orchestrator that dynamically delegates subtasks to either a conventional GUI Operator or a specialized Programmer agent, which can write and execute Python or Bash scripts. This hybrid approach allows the agent to bypass inefficient GUI action sequences for tasks like file management and data processing, while still leveraging visual interaction when necessary. We evaluate our system on the challenging OSWorld benchmark, where CoAct-1 achieves a new state-of-the-art success rate of 60.76%, significantly outperforming prior methods. Furthermore, our approach dramatically improves efficiency, reducing the average number of steps required to complete a task to just 10.15, compared to 15 for leading GUI agents. Our results demonstrate that integrating coding as a core action provides a more powerful, efficient, and scalable path toward generalized computer automation.

CoAct-1: Computer-using Agents with Coding as Actions

TL;DR

GUI-based agents struggle with long-horizon tasks due to brittle perception and inefficient action sequences. CoAct-1 introduces a hybrid multi-agent system that couples GUI manipulation with direct programmatic execution via a Programmer and an Orchestrator that dynamically assigns subtasks. It achieves a new OSWorld state-of-the-art 60.76% success rate (100-step budget) and an average of 10.15 steps per task, outperforming prior GUI-centric methods. The results show that incorporating coding as an action enhances robustness, efficiency, and scalability for general computer automation.

Abstract

Autonomous agents that operate computers via Graphical User Interfaces (GUIs) often struggle with efficiency and reliability on complex, long-horizon tasks. While augmenting these agents with planners can improve task decomposition, they remain constrained by the inherent limitations of performing all actions through GUI manipulation, leading to brittleness and inefficiency. In this work, we introduce a more robust and flexible paradigm: enabling agents to use coding as a enhanced action. We present CoAct-1, a novel multi-agent system that synergistically combines GUI-based control with direct programmatic execution. CoAct-1 features an Orchestrator that dynamically delegates subtasks to either a conventional GUI Operator or a specialized Programmer agent, which can write and execute Python or Bash scripts. This hybrid approach allows the agent to bypass inefficient GUI action sequences for tasks like file management and data processing, while still leveraging visual interaction when necessary. We evaluate our system on the challenging OSWorld benchmark, where CoAct-1 achieves a new state-of-the-art success rate of 60.76%, significantly outperforming prior methods. Furthermore, our approach dramatically improves efficiency, reducing the average number of steps required to complete a task to just 10.15, compared to 15 for leading GUI agents. Our results demonstrate that integrating coding as a core action provides a more powerful, efficient, and scalable path toward generalized computer automation.

Paper Structure

This paper contains 15 sections, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Computer-using Agent with Coding as Actions
  4. Multi-agent System Design for Computer Use
  5. Workflow and Memory Design
  6. Experiments
  7. Benchmark Dataset
  8. Baselines
  9. Implementation Details
  10. Results
  11. Discussion
  12. Efficiency Analysis
  13. CoAct-1 with different backbone
  14. Case Study
  15. Conclusions

Figures (3)

  • Figure 1: Multi-agent system design for our CoAct-1. This multi-agent system includes a programmer who can interact with the operating system through multi-round coding.
  • Figure 2: Illustration of CoAct-1 workflow. Given a user task, Orchestrator can choose either to call Programmer or GUI Operator to solve a subtask. A programmer can interact with the OS by coding, and a GUI Operator can interact with the OS by performing GUI operations.
  • Figure 3: CoAct-1 Efficiency and Step Modality Analysis.