AutoFlow: Automated Workflow Generation for Large Language Model Agents

Source: arXiv Authors: Zelong Li, Shuyuan Xu, Kai Mei, Wenyue Hua, Balaji Rama, Om Raheja, Hao Wang, He Zhu, Yongfeng Zhang Date: 2024-07-01 Primary category: cs.CL All categories: cs.CL, cs.AI, cs.LG

Abstract

AutoFlow proposes automatic generation of natural-language workflows for LLM agents, reducing dependence on manually authored procedures. It includes iterative workflow optimization and supports both fine-tuning-based and in-context generation modes, so the framework can work with open and closed models. The architectural signal is that workflows can be generated, inspected, and iteratively improved as explicit artifacts rather than remaining hidden inside one giant prompt.