Search papers, labs, and topics across Lattice.
This paper identifies the post-retrieval reuse of long-horizon agent trajectories as a critical bottleneck in trajectory memory and introduces a framework to evaluate this process. The authors propose query-conditioned reuse (QCR), which enhances the agent's ability to leverage past trajectories by providing a structured method to record and apply relevant procedures, achieving a 62.3% average success rate across various tasks while significantly reducing token usage. The findings reveal that QCR outperforms traditional full trajectory methods, particularly in scenarios with longer trajectories or changing contexts, thereby highlighting the importance of effective memory reuse in agent performance.
Query-conditioned reuse boosts agent success by 10.7 points while slashing token usage by nearly 50%, transforming how we leverage past experiences in AI tasks.
Retrieval can identify a past trajectory that may matter, yet it does not specify how an acting agent should use that trajectory after users, entities, constraints, or environment state have changed. We identify this post-retrieval reuse step as a distinct bottleneck for long-horizon trajectory memory and formulate an evaluation framework that holds candidate retrieval, target state, model, decoding, and tool budget fixed while varying the support delivered to the agent. We instantiate the framework with query-conditioned reuse (QCR), a deliberately simple target-bound note that records a reusable procedure, bindings to recover, applicability conditions, and verification requirements. QCR serves to test the reuse hypothesis rather than to claim a universally preferred memory format. Across 2,391 target instances in WebArena, WorkArena, and AppWorld, QCR reaches 62.3% average Success, 10.7 points above Full Trajectory, while using 48.9% fewer online tokens. Summary reranking selects a reusable memory for 94.8% of targets, placing end-task Success within 1.8 points of an oracle reusable selector. Analyses by trajectory length and source--target binding shift show that direct trajectory injection loses much of its utility as traces grow longer or source-specific values change, whereas target-bound support preserves a larger share of the measured gain. The resulting framework separates retrieval quality from the problem of turning retrieved experience into safe, useful support for a new task.
