ClickHouse in the Loop for Active Agents
Low-latency analytical systems can help active agents, but only when event loops include validation, context, and safety boundaries. For…
The archive
Writing, page 17
593 posts on Apache Iceberg, lakehouse architecture, data engineering and applied AI.
The Context Layer for AI Agents
A semantic layer is necessary, but agents also need lineage, quality, freshness, compliance, and ownership context. For data governance and…
Lakehouse as the Operating Layer for Agentic AI
Agentic AI announcements are useful when they validate the need for governed data, semantic context, and cost-aware execution. For data…
Event-Driven Table Compaction with Agents
Event-driven compaction is valuable when agents coordinate maintenance with workload signals, table health, and commit safety. For…
Fabric Agentic Analytics and Lakehouse Schema Design
Microsoft Fabric agentic analytics is a reminder that schemas, semantic models, and governed lakehouse design now shape AI behavior. For…
Fine-Grained Security for AI Agents
Machine-speed analytics requires machine-enforced policy, identity, masking, filtering, and audit controls. For security architects and…
Iceberg v4 Performance: Root Manifests and Calls
Apache Iceberg v4 discussion should focus on planning cost, metadata layout, and object storage round trips, not vague claims about faster…
What Is LTAP in the Lakehouse?
Lakehouse transactional analytical processing is useful only when teams define freshness, isolation, and workload boundaries clearly. For…
The Model Is Not the Moat
Enterprise AI advantage increasingly comes from governed context, semantic models, and operational data contracts, not only from model…
PyIceberg at Scale Without Apache Spark
Python-first Iceberg work is useful when it stays honest about what Python should and should not do. For Python data engineers and platform…