[Paper Review] ProAgent: From Robotic Process Automation to Agentic Process Automation

Yining Ye, Xin Cong+10 more|arXiv (Cornell University)|Nov 2, 2023

Robotic Process Automation Applications9 citations

TL;DR

The paper introduces Agentic Process Automation (APA) and instantiates ProAgent, an LLM-based agent system that constructs and executes workflows by coordinating specialized agents, moving beyond traditional RPA.

ABSTRACT

From ancient water wheels to robotic process automation (RPA), automation technology has evolved throughout history to liberate human beings from arduous tasks. Yet, RPA struggles with tasks needing human-like intelligence, especially in elaborate design of workflow construction and dynamic decision-making in workflow execution. As Large Language Models (LLMs) have emerged human-like intelligence, this paper introduces Agentic Process Automation (APA), a groundbreaking automation paradigm using LLM-based agents for advanced automation by offloading the human labor to agents associated with construction and execution. We then instantiate ProAgent, an LLM-based agent designed to craft workflows from human instructions and make intricate decisions by coordinating specialized agents. Empirical experiments are conducted to detail its construction and execution procedure of workflow, showcasing the feasibility of APA, unveiling the possibility of a new paradigm of automation driven by agents. Our code is public at https://github.com/OpenBMB/ProAgent.

Motivation & Objective

• Motivate overcoming RPA limitations in handling tasks requiring human-like intelligence and dynamic decision-making.
• Propose APA as a paradigm that offloads intelligent labor to LLM-based agents in both workflow construction and execution.
• Instantiate ProAgent to demonstrate agentic workflow construction and agentic workflow execution.
• Define an Agentic Workflow Description Language to formalize data flow and control flow for LLM understanding.
• Showcase feasibility through proof-of-concept experiments in a commercial scenario.

Proposed method

• Design Agentic Workflow Description Language using JSON for data flow and Python code for control flow.
• Introduce DataAgent and ControlAgent to handle data processing and dynamic decision-making during execution.
• Use GPT-4 for workflow construction with four iterative steps: action_define, action_implement, workflow_implement, task_submit.
• Implement workflow execution via a Python interpreter that runs mainWorkflow sequentially.
• Conduct proof-of-concept experiments on an n8n-based workflow platform, demonstrating construction and execution in Google Sheets–Slack/Email scenarios.

Experimental results

Research questions

• RQ1 Can LLM-based agents construct workflows autonomously from human instructions and orchestrate specialized agents for dynamic decision-making?
• RQ2 Is ProAgent capable of handling complex data processing and control-flow decisions within automated workflows?
• RQ3 What is the feasibility and practicality of Agentic Process Automation in a real-world workflow platform?
• RQ4 How can Agentic Workflow Description Language enable LLMs to generate consistent, executable workflows?

Key findings

• ProAgent demonstrates autonomous workflow construction from human input by generating a workflow language and Python-based control logic.
• DataAgent and ControlAgent enable dynamic data processing and conditional branching within workflows, improving flexibility.
• Proof-of-concept experiments show the feasibility of APA in constructing and executing agent-driven workflows on an open platform (n8n).
• The Agentic Workflow Description Language provides a structured, code-friendly format (JSON for data, Python for control) suited to LLM pretraining.
• The approach discusses integration with tool learning, process mining, and safety considerations, highlighting potential benefits and risks of agentic automation.