When Agents Disagree With Themselves: Measuring Behavioral Consistency in LLM-Based Agents

TL;DR

This paper investigates the inconsistency of LLM-based agents across repeated runs, revealing that behavioral variance correlates strongly with task failure and occurs early in the decision process, highlighting potential for early error detection.

Contribution

It provides the first large-scale analysis of behavioral consistency in LLM agents, linking early divergence to task failure and suggesting monitoring strategies for improved reliability.

Findings

Inconsistent behaviors occur in 2.0--4.2 action sequences per 10 runs.

High consistency correlates with 80--92% accuracy, low with 25--60%.

Most divergence happens at the first search step.

Abstract

Run the same LLM agent on the same task twice: do you get the same behavior? We find the answer is often no. In a study of 3,000 agent runs across three models (Llama 3.1 70B, GPT-4o, and Claude Sonnet 4.5) on HotpotQA, we observe that ReAct-style agents produce 2.0--4.2 distinct action sequences per 10 runs on average, even with identical inputs. More importantly, this variance predicts failure: tasks with consistent behavior ($\leq$2 unique paths) achieve 80--92% accuracy, while highly inconsistent tasks ($\geq$6 unique paths) achieve only 25--60%, a 32--55 percentage point gap depending on model. We trace variance to early decisions: 69% of divergence occurs at step 2, the first search query. Our results suggest that monitoring behavioral consistency during execution could enable early error detection and improve agent reliability.