GSA MCP Servers for Open Data AI Agents
The U.S. government publishes hundreds of thousands of datasets through Data.gov and agency portals, and almost none of them are shaped for…
The archive
Writing, page 12
593 posts on Apache Iceberg, lakehouse architecture, data engineering and applied AI.
Hybrid Lakehouse Design for Regulated Markets
A bank, a hospital network, and a defense contractor walk into a cloud migration and all three stop at the same wall: their data is not…
Iceberg v3 Deletion Vectors and Faster DML
Deleting one row from a data lake used to mean rewriting a whole file. If a 512 MB Parquet file held a million rows and you needed to…
Iceberg Variant Type for AI JSON Data
A single LLM response is not a single value. It carries the generated text, a reasoning trace, one or more tool calls with their arguments,…
Iceberg Concurrency for AI Agent Writes
A single autonomous agent can attempt more table commits in an hour than a team of analysts produces in a week. Multiply that by a fleet of…
MCP Stateless Gateways for Data Agents
A Model Context Protocol server that holds session state in memory is a server you cannot scale by adding replicas. The moment one process…
Policy Layer for Autonomous AI Data Agents
An autonomous agent can issue a thousand queries in the time it takes a human to read the results of one. That speed is the whole reason to…
REST Catalog Credential Vending for Iceberg
Handing a query engine a long-lived cloud storage key so it can read an Iceberg table is one of the most common and most dangerous patterns…
REST Catalog V2: Fixing Iceberg Protocol Debt
A single BI dashboard refresh can trigger dozens of catalog calls before a single byte of table data is scanned. An AI agent investigating…
Semantic View Autopilot for AI Governance
Most data glossaries are wrong by the time you read them. A column gets renamed, a metric changes its grain, a new product line ships, and…