Large Language Model-based Data Science Agent: A Survey

TL;DR

This survey addresses the challenge of deploying reliable LLM-based agents for data science by presenting a dual-perspective framework that links general agent design principles with concrete data-science workflows. It analyzes agent roles, execution structures, external knowledge sources, and reflection mechanisms, illustrating how single, dual, and multi-agent configurations support end-to-end DS tasks—from data preprocessing to visualization. The work identifies core design dimensions (roles, execution, knowledge, reflection) and surveys current benchmarks, highlighting limitations such as short-horizon tasks and data cleanliness considerations. It proposes three future directions—data-centric diagnostics, uncertainty-aware planning, and pipeline-level reflection—to improve reliability, interpretability, and robustness in DS agents. Overall, the paper offers a structured lens for building end-to-end, AI-assisted data science systems that integrate planning, verification, and data-quality checks into routine workflows.

Abstract

The rapid advancement of Large Language Models (LLMs) has driven novel applications across diverse domains, with LLM-based agents emerging as a crucial area of exploration. This survey presents a comprehensive analysis of LLM-based agents designed for data science tasks, summarizing insights from recent studies. From the agent perspective, we discuss the key design principles, covering agent roles, execution, knowledge, and reflection methods. From the data science perspective, we identify key processes for LLM-based agents, including data preprocessing, model development, evaluation, visualization, etc. Our work offers two key contributions: (1) a comprehensive review of recent developments in applying LLMbased agents to data science tasks; (2) a dual-perspective framework that connects general agent design principles with the practical workflows in data science.

Figures (14)

• Figure 1: Structure of This Survey
• Figure 2: Keyword cloud summarizing the most frequent concepts across the surveyed LLM-based data science agent papers. Larger words represent higher occurrence frequency. The colors in the cloud correspond to the major analysis dimensions discussed in this survey: red words represent agent role design (§\ref{['sec:analysis_llm:agent']}), blue words represent execution structure (§\ref{['sec:analysis_llm:execution']}), green words represent external knowledge integration (§\ref{['sec:analysis_llm:knowledge']}), and purple words represent reflection mechanisms (§\ref{['sec:analysis_llm:feedback']}).
• Figure 3: We illustrate the basic components for current data science agents: 1) agent role; 2) execution structure; 3) knowledge and 4) reflection.
• Figure 4: The basic structure for a single agent structure, with only the agent and execution environment.
• Figure 5: An example of planner and executor agent structure, where the planner generates a plan for the executor to execute in detail.