Researchagentsplanningweb automationbenchmarks
Agents Model Web Tasks As Planning Framework
|
7.6
A paper submitted to arXiv on March 13, 2026, by Rotem Dror and collaborators treats web automation tasks as sequential decision processes and maps modern agent architectures to classical planning paradigms. The authors introduce five trajectory-quality evaluation metrics and a dataset of 794 human-labeled WebArena trajectories, and compare Step-by-Step and Full-Plan-in-Advance agents, finding 38% human-aligned success for Step-by-Step and 89% element accuracy for Full-Plan.
Scoring Rationale
Strong methodology and new dataset drive score, limited by preprint status and narrower web-task scope.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
