Large Language Model Agents Are Not Always Faithful Self-Evolvers

TL;DR

This paper systematically investigates whether self-evolving LLM agents reliably depend on their experience, revealing a significant gap in their use of condensed experience and highlighting underlying causes.

Contribution

It provides the first comprehensive causal analysis of experience faithfulness in self-evolving LLM agents across multiple frameworks and environments.

Findings

Agents depend on raw experience but often ignore condensed experience.

The gap persists across different scales and configurations.

Underlying causes include semantic limitations, processing biases, and task regimes.

Abstract

Self-evolving large language model (LLM) agents continually improve by accumulating and reusing past experience, yet it remains unclear whether they faithfully rely on that experience to guide their behavior. We present the first systematic investigation of experience faithfulness, the causal dependence of an agent's decisions on the experience it is given, in self-evolving LLM agents. Using controlled causal interventions on both raw and condensed forms of experience, we comprehensively evaluate four representative frameworks across 10 LLM backbones and 9 environments. Our analysis uncovers a striking asymmetry: while agents consistently depend on raw experience, they often disregard or misinterpret condensed experience, even when it is the only experience provided. This gap persists across single- and multi-agent configurations and across backbone scales. We trace its underlying causes to three factors: the semantic limitations of condensed content, internal processing biases that suppress experience, and task regimes where pretrained priors already suffice. These findings challenge prevailing assumptions about self-evolving methods and underscore the need for more faithful and reliable approaches to experience integration.