Apache Iceberg Support Across the Major Hyperscalers
Every few weeks I get a version of the same question. A team has standardized on Apache Iceberg, they run most of their workloads on one…
The archive
Writing, page 8
593 posts on Apache Iceberg, lakehouse architecture, data engineering and applied AI.
Guardrails for Analytics Agents That Do More Than Answer Questions
An agent monitoring inventory levels notices a stockout risk, drafts a purchase order, and submits it. The logic was sound. The signal it…
Building Agent Telemetry Tables in Iceberg That Survive an Audit
An agent gives a customer-facing team a revenue number. Six weeks later someone asks where it came from. The application logs rolled off…
What Agentic Analytics Actually Costs, and How to Keep It Bounded
A data platform team gets a question from finance in month four of an agent rollout. The engine bill is up 38 percent and the model…
Running an Apache Iceberg Lakehouse With No Internet Connection
An engineer runs pip install pyiceberg on a classified network and it fails. There is no PyPI. There is no Maven Central, no Docker Hub, no…
When the Query Optimizer Starts Managing Its Own Materializations
A data engineer maintains 60 materialized views. Twelve of them accelerate queries nobody runs anymore, because the dashboard they served…
Why AI Agents Fail on Raw Data, and What to Give Them Instead
An analytics agent gets read access to the data lake. Someone asks it for last quarter's revenue by region. It finds a table named…
Why Iceberg V4 Wants to Retire Equality Deletes, and What Streaming Teams Should Do About It
A Flink job writes upserts into an Apache Iceberg table at a few thousand rows per second. The write side looks healthy. Checkpoints land…
The Five Layers Between Your Lakehouse and a Trustworthy Agent
An organization ships an analytics agent. It has access to the warehouse, a good model, and a well-written system prompt. Three weeks in,…
Apache Fluss and Kafka Solve Different Problems in an Iceberg Pipeline
A streaming team is asked to cut dashboard latency from six minutes to under thirty seconds. The pipeline is Kafka into Flink into Apache…