Query Routing at Machine Scale: Multi-Engine Workload Distribution for the Agentic Lakehouse
For most of the analytics era, query routing was a human problem with human solutions. Analysts learned which tool to open for which job,…
The archive
Writing, page 5
593 posts on Apache Iceberg, lakehouse architecture, data engineering and applied AI.
Securing the Agentic Lakehouse Gateway: A Threat Model for Prompt Injection, Exfiltration, and the Firewall That Reads Sentences
My article on building stateless tool gateways ended with a promise deferred: the gateway is where agent traffic converges, which makes it…
Semantic Layer Federation: One Meaning for Data That Lives Everywhere
Ask three systems in the same company what monthly recurring revenue was in July and you can get three answers, each computed correctly by…
Building Lightweight Serverless Ingestion to Apache Iceberg with PyIceberg and DuckDB
The most common ingestion job in most companies is small. A vendor drops a CSV in a bucket every hour. A webhook delivers a few thousand…
Building Stateless AI Tool Gateways with FastMCP, the 2026 MCP Spec, and Kubernetes
The gap between an agent demo and an agent platform is an infrastructure gap, and it has a specific shape. The demo runs one MCP server on…
Write-Audit-Publish with Apache Iceberg Branches: CI/CD for Your Data
Software engineering solved its most important quality problem decades ago with a single structural idea, applied everywhere and questioned…
Zero-Copy Warehouse Modernization: Moving to Apache Iceberg Without Downtime
Warehouse migrations have a reputation, and the reputation is earned. The classic project copies everything: export the tables, rebuild the…
The Plan and the Worker: Two Open Specifications for Agent Harnesses
Every agent harness solves the same two problems, and almost every one of them solves both privately. The first problem is decomposition. A…
Apache Arrow Flight and ADBC, and Why Database Connectivity Finally Went Columnar
A data scientist runs a query against a warehouse. The engine finishes the scan in three seconds. Then the notebook sits there for four…
Budgeting for Agentic Analytics When Every Question Costs Something Different
The pilot ran for six weeks with forty users and cost less than a team lunch. Someone approved rolling it out to eight hundred people. The…