Agent-E: From Autonomous Web Navigation to Foundational Design Principles in Agentic Systems
Keywords: Web Automation, Autonomous Agents, Self-Improvement, Hierarchical Architecture
Abstract: Web agents that can automate complex and monotonous tasks are becoming essential in streamlining workflows. Due to the difficulty of long-horizon planning, abundant state spaces in websites, and their cryptic observation space (i.e. DOMs), current web agents are still far from human-level performance. In this paper, we present a novel web agent, Agent-E \footnote. This agentic system introduces several architectural improvements over prior state-of-the-art web agents, such as hierarchical architecture, self-refinement, flexible DOM distillation and denoising method, and \textit{change observation} to guide the agent towards more accurate performance. Our Agent-E system without self-refinement achieves SOTA results on the WebVoyager benchmark, beating prior text-only benchmarks by over 20.5\% and multimodal agents by over 16\%. Our results indicate that adding a self-refinement mechanism can provide an additional 5.9\% improvement on the Agent-E system without self-refinement. We then synthesize our learnings into general design principles for developing agentic systems. These include the use of domain-specific primitive skills, the importance of distillation and de-noising of complex environmental observations, and the advantages of a hierarchical architecture.
Primary Area: applications to robotics, autonomy, planning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3884
Loading