Illuminating LLM Coding Agents: Visual Analytics for Deeper Understanding and Enhancement

TL;DR

This paper presents a visual analytics system for analyzing large language model-based coding agents, enabling detailed comparison and understanding of their debugging, process, and behavior across different models to improve development and debugging efficiency.

Contribution

The paper introduces a novel visual analytics system tailored for examining LLM coding agents, supporting multi-level analysis to enhance understanding and debugging of these agents.

Findings

System enables detailed code evolution analysis

Facilitates comparison of different solution processes

Provides insights into LLM behavioral variations

Abstract

Coding agents powered by large language models (LLMs) have gained traction for automating code generation through iterative problem-solving with minimal human involvement. Despite the emergence of various frameworks, e.g., LangChain, AutoML, and AIDE, ML scientists still struggle to effectively review and adjust the agents' coding process. The current approach of manually inspecting individual outputs is inefficient, making it difficult to track code evolution, compare coding iterations, and identify improvement opportunities. To address this challenge, we introduce a visual analytics system designed to enhance the examination of coding agent behaviors. Focusing on the AIDE framework, our system supports comparative analysis across three levels: (1) Code-Level Analysis, which reveals how the agent debugs and refines its code over iterations; (2) Process-Level Analysis, which contrasts different solution-seeking processes explored by the agent; and (3) LLM-Level Analysis, which highlights variations in coding behavior across different LLMs. By integrating these perspectives, our system enables ML scientists to gain a structured understanding of agent behaviors, facilitating more effective debugging and prompt engineering. Through case studies using coding agents to tackle popular Kaggle competitions, we demonstrate how our system provides valuable insights into the iterative coding process.