Field notes on enterprise data

Demand forecasting for brands: why integrated data has to come first

Why a demand forecast is only as good as the history behind it, what integrated data means for brand-led companies (beyond one big server), and what to fix before you trust a model.

The semantic layer: what it is and how to build it

What a semantic layer is in practice, how Iceberg and a catalog fit together, how DataHub and OpenLineage unify lineage, and a simple phased rollout for platform teams.

AI agents, memory, and databases: the shared-knowledge gap and why the semantic layer matters

Why chat memory isn’t organizational memory, how fragmented databases break agent answers, why a semantic layer matters, why it’s hard to build, and how teams fix it in practice.

A document-centric knowledge base: authority, metadata, and when to skip RAG

What a knowledge base is for, when organizations need one, and how versioned docs with structured metadata can ground programs—and agents—without vector retrieval as the core.

The enterprise AI agent problem is data, not the model

The limiting factor is usually data, not the model: why integration and clear definitions matter more than a bigger model.

Analytics agents in the enterprise: what actually works

Why chat-with-data isn’t enough, how to ground agents in trusted metrics, and the guardrails real programs use.

Why we're building this—and what breaks in enterprise data

The problem isn't missing tools; it's fragmented definitions, integrations, and trust. Here's why we're focused on coherence.