Agentic Data Stack Weekly #1

Welcome to the first issue of Agentic Data Stack Weekly — a curated digest of the most important developments at the intersection of AI agents and data infrastructure. We focus on what matters to data engineers and data team leaders building the next generation of data systems.

Top Stories

MCP Protocol Adoption Hits Inflection Point in Data Tooling

The Model Context Protocol (MCP) is rapidly becoming the standard interface between AI agents and data tools. This week, three major data platforms announced MCP server support, bringing the total count of MCP-enabled data tools past 40. The trend is clear: data tools that don't offer agent-friendly interfaces risk irrelevance.

• Source: modelcontextprotocol.io
• Why it matters: MCP standardizes how agents discover, query, and manipulate data tools — eliminating custom integration code and enabling truly portable data agents.

Apache Iceberg: The Table Format Agents Actually Understand

Apache Iceberg's REST catalog specification is emerging as the preferred interface for data agents. Unlike Hive metastore's JDBC dependency, Iceberg's REST API lets agents discover schemas, time-travel through data, and manage table evolution through simple HTTP calls.

• Source: iceberg.apache.org
• Why it matters: Agents need programmatic, stateless access to data catalogs. Iceberg's REST-first approach is purpose-built for this pattern.

Tools & Releases

DuckDB 1.2 Adds Agent-Friendly Extensions

DuckDB's latest release includes improved programmatic APIs and a new extension framework that makes it easier for agents to load and query heterogeneous data sources. The embedded SQL engine is becoming the go-to choice for agent-local data processing.

Unity Catalog Expands Multi-Engine Support

Databricks' open-source Unity Catalog now supports Apache Spark, Trino, and DuckDB as first-class query engines, making it a viable candidate for multi-engine agentic architectures.

Worth Reading

Why Semantic Layers Are the Missing Piece for Data Agents

Data agents that query raw tables without understanding business context produce unreliable results. Semantic layers (Cube, dbt Semantic Layer, MetriQL) provide the business logic and metric definitions that agents need to generate trustworthy analytics. Without them, agents are just fancy SQL generators.

The Data Team Leader's Guide to Agent Readiness

A practical framework for evaluating your data stack's readiness for AI agents. Key areas: catalog accessibility (can agents discover your data?), API coverage (can agents operate your tools programmatically?), and semantic clarity (do agents understand your business context?).

Community

Upcoming: Agentic Data Stack Meetup — March 28

Our next community session focuses on real-world patterns for integrating MCP servers with existing data platforms. Featuring case studies from teams running agents in production.

• Register: lu.ma/AgenticDataStack

---

Curated by the Agentic Data Stack community. Subscribe for weekly updates.