The Five Layers of an Agentic Lakehouse and Where the MCP Server Sits
Someone on your team connects an AI desktop client to a query engine, asks a question about last quarter, and gets an answer in fifteen…
The archive
Writing, page 6
593 posts on Apache Iceberg, lakehouse architecture, data engineering and applied AI.
Autonomous Table Optimization When Your Query Workload Stops Being Predictable
Table maintenance used to be a scheduling problem. You knew which tables were hot, you knew that dashboards filtered on order date and…
Building Apache Iceberg Lakehouses That Run Without an Internet Connection
A hospital system wants a lakehouse. Their security review says patient data never leaves facilities they control. A defense contractor…
Wiring Analytical Queries to Transactional APIs in Closed-Loop Decision Agents
An agent reads a table, decides something, and calls an API that changes the world. That sentence contains a distributed systems problem…
Surviving Optimistic Commit Collisions When Hundreds of Agents Write to Iceberg
The read path got all the attention. Agents query tables, a semantic layer keeps them honest, a catalog decides what they can see. That…
How Iceberg V3 Deletion Vectors Fixed Merge-on-Read for Streaming Tables
A change data capture pipeline lands updates from an operational database every thirty seconds. Under Iceberg V2 with merge-on-read, each…
How Iceberg V3 Variant Shredding Changed Semi-Structured Data on S3 Tables
Every data engineer has inherited the same table. It has four or five real columns and one column called payload, raw, body, or eventjson.…
Reading the Apache Iceberg V4 Proposals Before They Land
A Flink job commits every five seconds. Each commit writes one small Parquet file. It also writes a manifest, rewrites a manifest list, and…
Building an Honest TCO Model for Open Lakehouses and Proprietary Warehouses
Someone in finance forwards the cloud data warehouse invoice with a one-line question: why is this number growing faster than our data. An…
Why Agentic AI Needs a Governed Semantic Layer Behind the Model Context Protocol
An executive asks an AI assistant what revenue looked like last quarter. The assistant writes SQL against the warehouse, sums an amount…