AI summaryⓘ
The authors study how AI agents use memory during long tasks that involve many steps, like browsing the web or using software. Instead of just focusing on what memory to store or retrieve, they look at how agents actually use retrieved memory when deciding actions over time. They develop a framework called Entry–Propagation–Recovery (E-P-R) to analyze where memory causes changes in behavior, how mistakes spread, and how agents recover from errors. They find that agents often make mistakes early when they first use wrong memory, which then leads to repeated errors, making recovery difficult and hurting performance. The authors suggest that evaluating these AI agents should consider not just what memory is retrieved, but how it is used throughout a task.
AI agentsmemory augmentationlong-horizon tasksmulti-step action trajectoryEntry-Propagation-Recovery (E-P-R)WebArenaMemTrapBenchmemory retrievalagent complianceerror propagation
Abstract
Memory is becoming a core component of long-horizon AI agents, allowing agents to reuse past experience when operating web browsers, software tools, and other interactive environments. Existing work mostly treats memory as a supply problem, asking what experience to write, how to store it, and which entry to retrieve for the next task. Yet we still lack a clear account of how models consume retrieved memory across a multi-step action trajectory. This consumption process matters because it determines not only what memories should be retrieved, but also what models and control policies are needed to use them safely. To diagnose this process, we propose Entry--Propagation--Recovery (E-P-R), a trajectory-level framework that asks where memory first changes an action, whether that change carries forward, and whether the agent can recover after leaving a correct path. We instantiate E-P-R on WebArena and on MemTrapBench, a controlled benchmark we build to isolate these phases. We find that the main failure often begins at entry: agents adopt conflicting memory at the first exposed decision point even when it is task-wrong. Repeated exposure then amplifies this early error, while recovery after divergence is weak. Together, these effects create a compliance trap: across models, conflicting memory induces similar compliance rates, but once agents comply, their success rates collapse to a low floor. Stronger agents therefore suffer larger absolute damage because each compliance event erases more baseline capability. These results suggest that memory-augmented agents should be evaluated not only by retrieval quality or final success rate, but by how they consume memory throughout the trajectory.