ViqusViqus
Navigate
Company
Blog
About Us
Contact
System Status
Enter Viqus Hub

Coding Agents Get a Smart Upgrade: CoAct-1 Promises Automation Efficiency

Artificial Intelligence Automation Computer Vision Code Generation Salesforce GUI Automation Agent Technology LLM
August 12, 2025
Viqus Verdict Logo Viqus Verdict Logo 9
Code Reinvented
Media Hype 7/10
Real Impact 9/10

Article Summary

A team from Salesforce and USC has unveiled CoAct-1, a groundbreaking agent system designed to revolutionize computer automation. The system addresses the inherent brittleness of traditional GUI-based agents, which often struggle with complex, multi-step workflows. CoAct-1 operates as a three-agent team – an Orchestrator, a Programmer, and a GUI Operator – that intelligently combines the intuitive human-like strengths of GUI manipulation with the precision and efficiency of code execution. The Orchestrator plans and delegates tasks, while the Programmer leverages LLMs to generate and execute Python or Bash scripts. The GUI Operator handles visual interactions, using a VLM-based agent to navigate interfaces and click buttons. Critically, the system’s iterative workflow – receiving updates and screenshots after each step – allows for dynamic adjustment and error mitigation. Initial benchmarks on the OSWorld benchmark demonstrate CoAct-1’s superiority, achieving a 60.76% success rate compared to leading GUI-only agents, and completing tasks in just 10.15 steps on average. This advancement is particularly impactful for enterprise applications involving multi-tool workflows where full API access isn’t always available, like customer support or sales automation. While the technology presents significant potential, researchers emphasize that human oversight remains crucial. Future development includes refining the agent's reasoning capabilities and ensuring effective integration within broader enterprise systems. The team is also actively exploring real-world applications, with Salesforce citing customer support as a key initial target.

Key Points

  • CoAct-1 combines GUI manipulation with code execution for more reliable automation.
  • The system utilizes a three-agent team – Orchestrator, Programmer, and GUI Operator – to intelligently manage tasks.
  • An iterative workflow with dynamic adjustment based on system updates significantly reduces errors and improves task completion rates.

Why It Matters

This research represents a significant step toward truly robust and scalable agent automation. Traditional GUI-based agents, while convenient, are fundamentally limited by their reliance on imprecise human-like interactions. CoAct-1's ability to ‘translate’ complex workflows into executable code opens up possibilities for automating a wider range of tasks, particularly in enterprise environments where integration with diverse systems and the need for greater reliability are paramount. For business leaders, this translates to increased operational efficiency, reduced human error, and the potential to unlock new levels of automation previously unattainable with conventional agent technologies. The move away from brittle, mouse-click reliant systems has the potential to create a more sustainable and efficient path for AI deployment.

You might also be interested in