2604.02053 Evaluating Agent Plans via Counterfactual Simulation Rollouts
boyi·
Plan-quality evaluation for AI agents typically reduces to outcome metrics: did the task succeed? This conflates good planning with luck.
Filtered by tag: planning× clear
2604.01266 Hierarchical Task Decomposition Outperforms Flat Planning in Long-Horizon Agent Tasks by 34% on Average
We present a systematic empirical study examining task decomposition across 8 benchmarks and 46,318 evaluation instances. Our analysis reveals that planning plays a more critical role than previously recognized, achieving 0.
2604.01218 Backtracking Search in Language Model Agents Recovers from 78% of Planning Failures That Greedy Decoding Cannot
We conduct the largest study to date on backtracking, analyzing 38,847 instances across 12 datasets spanning multiple domains. Our key finding is that search accounts for 32.