Exploring Autonomous Agents: A Closer Look at Why They Fail When Completing Tasks

2025-08-18

AI Summary

Existing evaluation of LLM-based autonomous agents overemphasizes task success rates and lacks systematic failure attribution. Method: We introduce a benchmark comprising 34 programmable tasks and propose a fine-grained failure attribution taxonomy spanning planning, execution, and response generation stages. Leveraging three open-source agent frameworks and two LLM families, we conduct in-depth error analysis. Contribution/Results: Experiments reveal an overall task completion rate of ~50%, with failures predominantly stemming from suboptimal planning and insufficient self-diagnosis capability. We accordingly propose mitigation strategies—including stepwise verification, explicit state tracking, and reflective response generation. This work establishes the first reproducible, fine-grained failure attribution framework for agent interaction processes and inter-component communication, providing both an empirically grounded evaluation paradigm and actionable guidance for building robust autonomous agents.

Technology Category

• Multiagent Systems: Multiagent Planning
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Autonomous agent systems powered by Large Language Models (LLMs) have demonstrated promising capabilities in automating complex tasks. However, current evaluations largely rely on success rates without systematically analyzing the interactions, communication mechanisms, and failure causes within these systems. To bridge this gap, we present a benchmark of 34 representative programmable tasks designed to rigorously assess autonomous agents. Using this benchmark, we evaluate three popular open-source agent frameworks combined with two LLM backbones, observing a task completion rate of approximately 50%. Through in-depth failure analysis, we develop a three-tier taxonomy of failure causes aligned with task phases, highlighting planning errors, task execution issues, and incorrect response generation. Based on these insights, we propose actionable improvements to enhance agent planning and self-diagnosis capabilities. Our failure taxonomy, together with mitigation advice, provides an empirical foundation for developing more robust and effective autonomous agent systems in the future.

Problem

• Analyzing failure causes in autonomous agent task completion
• Evaluating agent frameworks and LLM backbones systematically
• Proposing improvements for agent planning and self-diagnosis

Innovation

• Benchmark with 34 tasks for agent assessment
• Three-tier taxonomy for failure causes analysis
• Improvements for agent planning and self-diagnosis