More Memory, Worse Agents: Error Reproduction and Anti-Persistence in LLM Agents
Ruhan Wang and Dongruo Zhou (Indiana University) with Kishan Panaganti (Tencent) identify error reproduction, in which wrong rules stored in agent memory regenerate through new trajectories, and propose Anti-Persistence, a memory design that stores only trajectories and outcomes and rebuilds guidance on demand.
Ask this paper
Error reproduction. A faulty stored abstraction shapes the next execution, that execution becomes the basis for new abstractions, and the error is derived again, so it persists even after the original entry is removed.
Anti-Persistence. The agent keeps trajectories and outcomes. An adaptive selector uses prior outcomes to decide whether and how to synthesize task-specific guidance, and the guidance is discarded after execution.
Results. Across benchmarks and backbones, Anti-Persistence has the highest overall held-out success and the lowest average LLM-call cost among the memory designs evaluated.
Design rule. The authors recommend separating the accumulation of experience from the persistence of abstractions in memory-based self-improvement.
Abstract
LLM agents can improve across tasks without updating model parameters by storing experience from their own executions in external memory and reusing it to guide later decisions. A central design question is whether to retain only observed trajectories and outcomes or also the rules, summaries, and skills inferred from them. Retaining these abstractions allows agents to reuse strategies without repeatedly deriving them from raw trajectories. However, errors in stored abstractions can propagate to subsequent decisions through retrieval and reuse. They can also shape execution trajectories, which then become the basis for new abstractions that reproduce the original errors. We term this feedback mechanism error reproduction: errors are regenerated through a cycle of abstraction, execution, and re-abstraction, extending their persistence beyond direct reuse of the original faulty knowledge. We propose Anti-Persistence, a memory design that accumulates trajectories and outcomes while reconstructing task-specific guidance on demand. An adaptive selector uses prior outcomes to decide whether and how to synthesize guidance, which is discarded after execution. This prevents its direct reuse as a stored object, although indirect feedback through retained trajectories remains possible. Across different benchmarks and backbones, Anti-Persistence achieves higher overall held-out success and the lowest average LLM-call cost among the evaluated memory designs. These results support separating experience accumulation from abstraction persistence in memorybased self-improvement.