Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

WorkTeam: Constructing Workflows from Natural Language with Multi-Agents

Hanchao Liu, Rongjun Li, Weimin Xiong, Ziyu Zhou, Wei Peng

TL;DR

WorkTeam introduces a multi-agent NL2Workflow framework (supervisor, orchestrator, filler) to overcome the limitations of single-LLM approaches for enterprise workflow construction. The orchestrator handles component filtering and arrangement, while the filler fills in component parameters via templates and LLM-based generation, all coordinated by the supervisor. A new HW-NL2Workflow dataset with 3,695 real-world workflows supports training and evaluation. Empirical results on HW-NL2Workflow show substantial gains over baselines, achieving EMR $52.7\%$, AA $88.9\%$, and PA $73.2\%$, demonstrating improved accuracy and stability for enterprise NL2Workflow services. The work provides a practical, scalable path toward robust NL2Workflow systems in real-world settings.

Abstract

Workflows play a crucial role in enhancing enterprise efficiency by orchestrating complex processes with multiple tools or components. However, hand-crafted workflow construction requires expert knowledge, presenting significant technical barriers. Recent advancements in Large Language Models (LLMs) have improved the generation of workflows from natural language instructions (aka NL2Workflow), yet existing single LLM agent-based methods face performance degradation on complex tasks due to the need for specialized knowledge and the strain of task-switching. To tackle these challenges, we propose WorkTeam, a multi-agent NL2Workflow framework comprising a supervisor, orchestrator, and filler agent, each with distinct roles that collaboratively enhance the conversion process. As there are currently no publicly available NL2Workflow benchmarks, we also introduce the HW-NL2Workflow dataset, which includes 3,695 real-world business samples for training and evaluation. Experimental results show that our approach significantly increases the success rate of workflow construction, providing a novel and effective solution for enterprise NL2Workflow services.

WorkTeam: Constructing Workflows from Natural Language with Multi-Agents

TL;DR

WorkTeam introduces a multi-agent NL2Workflow framework (supervisor, orchestrator, filler) to overcome the limitations of single-LLM approaches for enterprise workflow construction. The orchestrator handles component filtering and arrangement, while the filler fills in component parameters via templates and LLM-based generation, all coordinated by the supervisor. A new HW-NL2Workflow dataset with 3,695 real-world workflows supports training and evaluation. Empirical results on HW-NL2Workflow show substantial gains over baselines, achieving EMR , AA , and PA , demonstrating improved accuracy and stability for enterprise NL2Workflow services. The work provides a practical, scalable path toward robust NL2Workflow systems in real-world settings.

Abstract

Workflows play a crucial role in enhancing enterprise efficiency by orchestrating complex processes with multiple tools or components. However, hand-crafted workflow construction requires expert knowledge, presenting significant technical barriers. Recent advancements in Large Language Models (LLMs) have improved the generation of workflows from natural language instructions (aka NL2Workflow), yet existing single LLM agent-based methods face performance degradation on complex tasks due to the need for specialized knowledge and the strain of task-switching. To tackle these challenges, we propose WorkTeam, a multi-agent NL2Workflow framework comprising a supervisor, orchestrator, and filler agent, each with distinct roles that collaboratively enhance the conversion process. As there are currently no publicly available NL2Workflow benchmarks, we also introduce the HW-NL2Workflow dataset, which includes 3,695 real-world business samples for training and evaluation. Experimental results show that our approach significantly increases the success rate of workflow construction, providing a novel and effective solution for enterprise NL2Workflow services.

Paper Structure

This paper contains 30 sections, 4 equations, 12 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Natural Language to Workflow
  4. Multi-Agents
  5. Methods
  6. The Supervisor Agent
  7. The Orchestrator Agent
  8. Component Filtering Tool
  9. Component Orchestration Tool
  10. The Filler Agent
  11. Template Lookup Tool
  12. Parameter Filling Tool
  13. HW-NL2Workflow
  14. Data Statistics
  15. Component Resources
  16. ...and 15 more sections

Figures (12)

  • Figure 1: An example of generating workflows (JSON format) from text instruction.
  • Figure 2: The overall architecture of the proposed WorkTeam framework.
  • Figure 3: Examples in the component set $C$ of HW-NL2Workflow.
  • Figure 4: Examples in the component parameter description set $T_{desc}$ of HW-NL2Workflow.
  • Figure 5: Examples in the blank parameter template set $T_{blank}$ of HW-NL2Workflow.
  • ...and 7 more figures