Agent Workflow Memory

Source: arXiv Authors: Zora Zhiruo Wang, Jiayuan Mao, Daniel Fried, Graham Neubig Date: 2024-09-11 Primary category: cs.CL All categories: cs.CL

Abstract

Agent Workflow Memory induces reusable workflows from past trajectories and selectively reuses them to guide later agent generations. The method works in both offline and online settings and improves long-horizon web-navigation performance by learning routines that can be recalled rather than rediscovered. This is one of the clearest examples of workflow artifacts becoming an external memory substrate rather than a one-shot planning output.