The Data Agent Stack - Part 5: Tools, Query Execution, and the Analyst Loop

2026-02-17 · Source: The Agent Stack · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, long

Summary

This article, part 5 of "The Data Agent Stack" series, details the critical role of tool curation and safe execution in building effective data agents, emphasizing that "more tools are not always more capability." It highlights that data agents, unlike human analysts, struggle with overlapping or unclear tool functionality, leading to incorrect answers. Drawing lessons from OpenAI, ByteByteGo, and Spotify, the article advocates for a curated tool surface with clear authority, distinguishing between metadata lookup, warehouse execution, runtime context, orchestration, code search, knowledge retrieval, publishing, and workflow tools. It also stresses the importance of an "analyst loop" (observe, act, verify, revise) over a linear text-to-SQL process, incorporating features like clarifying questions, sensible defaults, interrupts, and versioned workflows. Furthermore, the article details essential safety measures for SQL execution, including validation, cost controls, and transparent evidence linking.

Key takeaway

For AI/MLOps Engineers building data agents, prioritize a highly curated tool surface with explicit authority and robust execution safety. You should implement an iterative "analyst loop" that includes validation, cost controls, and visible assumptions. This approach prevents agents from making expensive errors or publishing unverifiable results, ensuring trustworthy and auditable data analysis. Version reusable workflows and support user interrupts to enhance usability and reliability.

Key insights

Data agents require a curated, high-authority tool surface and an iterative analyst loop for reliable, safe, and verifiable results.

Principles

• "Less is more" for data agent tools.
• Tools are capability contracts, not just API wrappers.
• Analysis is an iterative loop, not a straight line.

Method

A data agent should follow an "analyst loop": plan, inspect, execute, observe, validate, revise, and explain, interleaving reasoning with tool actions.

In practice

• Curate tools by authority, not organizational structure.
• Implement safety checks directly in the SQL execution path.
• Make all agent defaults and assumptions visible to users.

Topics

• Data Agents
• Tool Curation
• SQL Execution Safety
• Analyst Loop
• LLM Tooling
• Data Governance

Best for: AI Engineer, MLOps Engineer, Data Scientist

Related on AIssential

• The Data Agent Stack - Part 1: What Is a Data Agent? — A data agent is a governed analysis loop that proves answers, not just generates SQL, by integrating context and validation.
• Agent Loop Engineering: How to Build Reliable AI Agents for Production — Reliable AI agents require code-enforced decision loops, not just LLM interpretation, to prevent uncontrolled behavior.
• Build an Agent That Thinks Like a Data Scientist: How We Hit #1 on DABStep with Reusable Tool Generation — Separating knowledge building from inference via reusable tool generation dramatically improves agent performance and efficiency.
• Designing and Building an AI DataOps Incident Agent — An AI multi-agent system automates DataOps incident investigation, root cause analysis, and resolution planning.
• How OpenAI Built Its Data Agent — The reliability of AI agents at scale stems from strong data foundations and carefully engineered context, not complex agent architectures.
• Production-Grade Text-to-SQL Agent with Claude Code, LangGraph, Langfuse, FastAPI and Qdrant — Building robust Text-to-SQL agents requires RAG, explicit state machines, and rigorous evaluation to prevent hallucinations and ensure reliability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Agent Stack.