Curated papers and resources on Data Agents — from the perspective of proposed autonomy levels. Companion list for our (incoming) survey.
If you find this repository useful or inspiring, please kindly give us a star ⭐️ and cite our survey:
Coming soon
The way humans interact with data is undergoing a profound transformation. Data agents — LLM-powered systems designed to orchestrate the Data + AI ecosystem — are emerging as a promising solution for automating and democratizing data-related tasks across its lifecycle, from management and preparation to analysis.
However, the term "data agent" is currently used inconsistently across research and industry, resulting in considerable ambiguity. Systems with vastly different capabilities in autonomy, reliability, and task complexity are often labeled the same way. This creates a "Babel Tower" crisis where mismatched expectations and unclear accountability threaten to undermine user trust and impede healthy development of the field.
This repository — a companion to our survey — introduces a layered taxonomy (L0-L5) for data agents based on their degree of autonomy, providing a common framework to clarify capability boundaries and lines of accountability at each level.
As mentioned above, to bring clarity to the diverse landscape of data agents, we propose a layered taxonomy based on their degree of autonomy. This framework maps the progressive shift of responsibility from human to agent, defining the distinct roles each plays at every stage, as summarized in the overview figure and the table below.
| Level | Degree of Autonomy | Human Role | Data Agent Role |
|---|---|---|---|
| L0 | Manual/No Autonomy | Dominator (Solo) | N/A (None) |
| L1 | Assisted | Dominator (Editing) | Assistant (Auxiliary) |
| L2 | Partial Autonomy | Dominator (Orchestrating) | Executor (Procedural) |
| L3 | Conditional Autonomy | Supervisor (Overseeing) | Dominator (Autonomous) |
| L4 | High Autonomy | Onlooker (Delegating) | Dominator (Proactive) |
| L5 | Full Autonomy | N/A (None) | Dominator (Generative) |
The transition between these levels represents more than just incremental progress; each step up the hierarchy requires a significant evolutionary leap as shown below. These leaps involve fundamental shifts in a data agent's capabilities—such as gaining environmental perception (L1→L2), achieving autonomous orchestrating and dominating the task (L2→L3), attaining proactive self-governance with supervision removed (L3→L4), or pioneering new paradigm (L4→L5).
We index papers by autonomy level, then by data-related tasks across Data Management, Data Preparation, and Data Analysis. Most existing work clusters in L1–L3, L4–L5 are aspirational. We also list relevant surveys, tutorials and benchmarks.
In L0 level, data-related tasks are performed entirely by human experts without any automation. The process is completely human-driven, requiring extensive domain knowledge and solid technical expertise, making it highly specialized and time-consuming.
At L1 level, data agents start to have the capabilities to provide preliminary and single-point assistance through typical question-answering interactions. While they can help with atomic tasks like code peices generation, they lack environmental perception and require considerable human validation, editing, and optimization.
- Query Rewriting via LLMs — arXiv
- Query Performance Explanation through LLM for HTAP Systems — arXiv
- Demonstration of DB-GPT: Next Generation Data Interaction System Empowered by Large Language Models — VLDB'24 Demo
- LLMClean: Context-Aware Tabular Data Cleaning via LLM-Generated OFDs — arXiv
- HAIPipe: Combining Human-generated and Machine-generated Pipelines for Data Preparation — SIGMOD'23
-
BoostER: Leveraging Large Language Models for Enhancing Entity Resolution — WWW'24
-
Leveraging Large Language Models for Entity Matching — arXiv
-
Using ChatGPT for Entity Matching — ADBIS 2023
-
Cost-Effective In-Context Learning for Entity Resolution — ICDE'24
-
Schema Matching with Large Language Models: An Experimental Study — VLDB'24
-
Magneto: Scaling Schema Matching by Combining Small and Large Language Models — arXiv
-
Matchmaker: Self-Improving LLM Programs for Schema Matching — arXiv
-
LLMs4SchemaDiscovery: A Human-in-the-Loop Workflow for Scientific Schema Discovery using LLMs — arXiv
- Table Meets LLM: Can Large Language Models Understand Structured Table Data? A Benchmark and Empirical Study — WSDM'24
- TableLlama: Towards Open Large Generalist Models for Tables — NAACL'24
- Tab-CoT: Zero-shot Tabular Chain of Thought — ACL'23 Findings
- Are Large Language Models Good Statisticians? — NeurIPS'24
- Evaluating the Text-to-SQL Capabilities of Large Language Models — arXiv
- C3: Zero-shot Text-to-SQL with ChatGPT — arXiv
- DIN-SQL: Decomposed In-Context Learning of Text-to-SQL with Self-Correction — NeurIPS'23
- ACT-SQL: In-context learning for text-to-SQL with automatically-generated chain-of-thought — EMNLP'23 Findings
- Text-to-SQL Empowered by Large Language Models: A Benchmark Evaluation — VLDB'24
- Data2Vis: Automatic Generation of Data Visualizations from Natural Language — arXiv
- Automated Data Visualization from Natural Language via Large Language Models — SIGMOD'24
- Generating Analytic Specifications for Data Visualization from Natural Language Queries using Large Language Models — VIS'24 NLVIZ workshop
- HAIChart: Human and AI Paired Visualization System — VLDB'24
- REPLUG: Retrieval-Augmented Black-Box Language Models — NAACL'24
- M3DocRAG: Multi-modal Retrieval is What You Need for Multi-page Multi-document Understanding — arXiv
- ReportGPT: Human-in-the-Loop Verifiable Table-to-Text Generation — EMNLP'24 Industry
At this level, data agents gain environmental perception and tool-invocation capabilities, enabling them to execute bounded sub-tasks and multi-step procedures. While they can follow human-orchestrated workflows, the overall process is still dominated by human direction.
- ROMAS: A Role-Based Multi-Agent System for Database monitoring and Planning — arXiv
- GPTuner: A Manual-Reading Database Tuning System — VLDB'24
- D-Bot: Database Diagnosis System using Large Language Models — VLDB'24
- Automatic Database Configuration Debugging using Retrieval-Augmented Language Models — SIGMOD'25
- R-Bot: An LLM-based Query Rewrite System — VLDB'25
- QUITE: A Query Rewrite System Beyond Rules with LLM Agents — arXiv
- Revolutionizing Database Q&A with Large Language Models: Comprehensive Benchmark and Evaluation — KDD'25
- ChatPipe: Orchestrating Data Preparation Pipelines by Optimizing Human–ChatGPT Interactions — SIGMOD'24 Demo
- CleanAgent: Automating Data Standardization with LLM-based Agents — arXiv
- AutoDCWorkflow: LLM-based Data Cleaning Workflow Auto-Generation — arXiv
- Text-to-Pipeline: Bridging Natural Language and Data Preparation Pipelines — arXiv
- Weak-to-Strong Prompts with Lightweight-to-Powerful LLMs for High-Accuracy, Low-Cost, and Explainable Data Transformation — VLDB'24
- Exploring LLM Agents for Cleaning Tabular Machine Learning Datasets — ICLR'25 Workshop on Foundation Models in the Wild
- AutoPrep: Natural Language Question-Aware Data Preparation with a Multi-Agent Framework — VLDB'25
- Match, Compare, or Select? An Investigation of Large Language Models for Entity Matching — COLING'25
- Agent-OM: Leveraging LLM Agents for Ontology Matching — VLDB'25
- Knowledge Graph-based Retrieval-Augmented Generation for Schema Matching — arXiv
- Binder: Binding Language Models in Symbolic Languages — ICLR'23
- StructGPT: A General Framework for Large Language Model on Structured Data — EMNLP'23
- ReAcTable: Enhancing ReAct for Table Question Answering (VLDB'24) — VLDB'24
- Table-Critic: A Multi-Agent Framework for Collaborative Criticism and Refinement in Table Reasoning — ACL'25
- Large Language Models are Versatile Decomposers: Decomposing Evidence and Questions for Table-based Reasoning — SIGIR'23
- Chain-of-Table: Evolving Tables in the Reasoning Chain for Table Understanding — ICLR'24
- AutoTQA: Towards Autonomous Tabular Question Answering through Multi-Agent Large Language Models — PVLDB'24
- TableMaster: A Recipe to Advance Table Understanding with Language Models — arXiv
- CHASE-SQL: Multi-Path Reasoning and Preference Optimization for Text-to-SQL — ICLR'25
- MAC-SQL: A Multi-Agent Collaborative Framework for Text-to-SQL — COLING'25
- Alpha-SQL: Zero-Shot Text-to-SQL using Monte Carlo Tree Search — arXiv
- ReFoRCE: A Text-to-SQL Agent with Self-Refinement, Consensus Enforcement, and Column Exploration — arXiv
- ChatBI: Towards Natural Language to Complex Business Intelligence SQL — arXiv
- MatPlotAgent: Method and Evaluation for LLM-Based Agentic Scientific Data Visualization — arXiv
- DeepVIS: Bridging Natural Language and Data Visualization Through Step-wise Reasoning — VIS'25
- C2: Scalable Auto-Feedback for LLM-based Chart Generation — NAACL'25
- PlotGen: Multi-Agent LLM-based Scientific Data Visualization via Multimodal Feedback — WWW'25
- nvAgent: Automated Data Visualization from Natural Language via Collaborative Agent Workflow — ACL'25
- DataNarrative: Automated Data-Driven Storytelling with Visualizations and Texts — EMNLP'24
- Multimodal DeepResearcher: Generating Text–Chart Interleaved Reports with an Agentic Framework — arXiv
- DAgent: A Relational Database-Driven Data Analysis Report Generation Agent — arXiv
- Semantic Operators: A Declarative Model for Rich, AI-Augmented Data Analytics — arXiv
- LAMBDA: A Large Model Based Data Agent — arXiv
- InfiAgent-DABench: Evaluating Agents on Data Analysis Tasks — ICML'24
Level L3 marks a critical transition where data agents evolve from procedural executors into autonomous directors for data-related tasks. At this stage, they can independently decompose high-level goals, orchestrate and optimize tailored, end-to-end pipelines, shifting the human to a supervisory role. While recent pioneering efforts show promise, they are largely considered very early-stage "Proto-L3" systems. Consequently, the pursuit of more autonomous, reliable, versitile and comprehensive L3 data agents remains a key objective in both academia and industry.
- DBAIOps: A Reasoning LLM-Enhanced Database Operation and Maintenance System using Knowledge Graphs — arXiv
- GaussMaster: An LLM-based Database Copilot System — arXiv
- Unify: A System For Unstructured Data Analytics — ICDE'25
- DataMosaic: Explainable and Verifiable Multi-Modal Data Analytics through Extract-Reason-Verify (arXiv'25) — arXiv
- Chat2Data: An Interactive Data Analysis System with RAG, Vector Databases and LLMs — VLDB'24 Demo
- Data Interpreter: An LLM Agent For Data Science — ACL'25 Findings
- AOP: Automated and Interactive LLM Pipeline Orchestration for Answering Complex Queries — CIDR'25
- Palimpzest: Optimizing AI-Powered Analytics with Declarative Query Processing — CIDR'25
- AgenticData: An Agentic Data Analytics System for Heterogeneous Data — arXiv
- SPIO: Ensemble and Selective Strategies via LLM-Based Multi-Agent Planning in Automated Data Science — arXiv
- Data Agent: A Holistic Architecture for Orchestrating Data+AI Ecosystems — arXiv
Data agents at L4 can achieve sustained self-governance with proactive monitoring and optimization across the data lifecycle. They can operate autonomously for extended periods without human supervision, actively providing insights and feedback while maintaining reliability.
The ultimate vision of fully autonomous data agents that can function as expert data scientists, capable of knowledge creation and paradigm innovation for data-related tasks.
- Natural Language Interfaces for Tabular Data Querying and Visualization: A Survey — TKDE'24
- A Survey of Text-to-SQL in the Era of LLMs: Where are we, and where are we going? — TKDE'25
- A Survey of LLM x DATA — arXiv
- Large Language Models for Data Science: A Survey — arXiv
- A Survey on Large Language Model-based Agents for Statistics and Data Science — arXiv
- Large Language Model-based Data Science Agent: A Survey — arXiv
- Large Language Models for Data Discovery and Integration: Challenges and Opportunities — IEEE Data Eng. Bull. 2025
- Large Language Models for Data Annotation and Synthesis: A Survey — EMNLP'24
- Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems — arXiv
- Data+AI: LLM4Data and Data4LLM — SIGMOD'25 Tutorial
- LLM for Data Management — VLDB'24 Tutorial