Search papers, labs, and topics across Lattice.
This paper introduces PlanAhead, a static planner-executor framework, to empirically evaluate the impact of different natural language plan representations (sequential subgoals, narrative, pseudocode, checklist) on the performance of LLM-based web agents. They categorize WebArena tasks into difficulty levels and use Achievement Rate (AR) and Solved-Task Consistency (STC) to evaluate plan representations across different LLMs. The study reveals that both the plan representation and the underlying LLM significantly influence web-agent robustness and task success on hard WebArena tasks.
How you represent a plan matters more than which LLM you use when building robust web agents.
Despite recent advances, LLM-based web agents still struggle with limited exploration, omission of critical steps, and sensitivity to task constraints. Prior work suggests that many of these failures stem from weaknesses in planning, yet the impact of alternative natural language plan representation remains unexplored. To address this, we introduce PlanAhead, a static planner-executor framework that evaluates the impact of plan representation in agent performance. We first automatically categorize WebArena tasks into 3 difficulty levels, enabling consistent difficulty grading without human annotation. Then we systematically evaluate 4 different plan representations on the tasks categorized as hard: sequential subgoals, narrative, pseudocode, and checklist; across different families of multimodal LLM powered agents (OpenAI, Alibaba, and Google). To account for stochastic variability, we introduce two novel evaluation metrics: Achievement Rate (AR) and Solved-Task Consistency (STC). Our results show that both, the plan formulation and the underlying LLM generating the plan, significantly influence web-agent robustness and task success.