WebOperator: Action-Aware Tree Search for Autonomous Agents in Web Environment
Mahir Labib Dihan, Tanzima Hashem, Mohammed Eunus Ali, Md Rizwan Parvez
2025-12-16
Summary
This paper introduces a new system called WebOperator that helps AI agents navigate the internet more effectively and reliably.
What's the problem?
Current AI agents that interact with websites tend to make decisions one step at a time, without thinking about the bigger picture or potential consequences. This is especially difficult on websites because you can only see part of the information at once, and a wrong click can lead to a complicated process of going back and trying again. Existing methods for planning ahead don't work well because they can cause unintended changes on the website or assume every action can be easily undone, which isn't always true.
What's the solution?
WebOperator uses a 'tree-search' approach, which means it explores different possible paths before committing to one. It carefully considers both how rewarding an action might be and how safe it is, and it checks to make sure going back to a previous step won't cause problems. It also generates a variety of possible actions to consider and filters out ones that won't work, ensuring a more robust exploration of the website.
Why it matters?
This research is important because it significantly improves the ability of AI agents to successfully complete tasks on the web. By allowing agents to plan ahead and safely backtrack, WebOperator achieves a much higher success rate on challenging web-based tasks, demonstrating the value of combining strategic thinking with careful execution.
Abstract
LLM-based agents often operate in a greedy, step-by-step manner, selecting actions solely based on the current observation without considering long-term consequences or alternative paths. This lack of foresight is particularly problematic in web environments, which are only partially observable-limited to browser-visible content (e.g., DOM and UI elements)-where a single misstep often requires complex and brittle navigation to undo. Without an explicit backtracking mechanism, agents struggle to correct errors or systematically explore alternative paths. Tree-search methods provide a principled framework for such structured exploration, but existing approaches lack mechanisms for safe backtracking, making them prone to unintended side effects. They also assume that all actions are reversible, ignoring the presence of irreversible actions-limitations that reduce their effectiveness in realistic web tasks. To address these challenges, we introduce WebOperator, a tree-search framework that enables reliable backtracking and strategic exploration. Our method incorporates a best-first search strategy that ranks actions by both reward estimates and safety considerations, along with a robust backtracking mechanism that verifies the feasibility of previously visited paths before replaying them, preventing unintended side effects. To further guide exploration, WebOperator generates action candidates from multiple, varied reasoning contexts to ensure diverse and robust exploration, and subsequently curates a high-quality action set by filtering out invalid actions pre-execution and merging semantically equivalent ones. Experimental results on WebArena and WebVoyager demonstrate the effectiveness of WebOperator. On WebArena, WebOperator achieves a state-of-the-art 54.6% success rate with gpt-4o, underscoring the critical advantage of integrating strategic foresight with safe execution.