Why Do LLM-based Web Agents Fail? A Hierarchical Planning Perspective
Mohamed Aghzal, Gregory J. Stein, Ziyu Yao · 2026
Read Why Do LLM-based Web Agents Fail? A Hierarchical Planning Perspective by Mohamed Aghzal, Gregory J. Stein, Ziyu Yao (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-14248.txt.
Description
Aghzal, Stein, and Yao ask why language model web agents fail at long web tasks, using a three-layer planning lens. The layers split high-level planning, low-level execution, and replanning. Structured PDDL plans beat plain language plans for focus and brevity, they find. Yet low-level execution, the clicking and grounding step, remains the main bottleneck. The authors argue that better perception and adaptive control matter more than grander plans. The framework gives a clear way to locate agent failures.
Catalog record
| CREATOR | Mohamed Aghzal, Gregory J. Stein, Ziyu Yao |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69836 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.14248