Case studies from the past year: a real-time data layer built from scratch for an LLM agent, a BI platform migration, the ClickHouse aggregation pattern that keeps showing up, and what happens when you shrink 50 agents down to skills.
-
A data model built for dashboards and one built for an AI agent have fundamentally different jobs. Here's why I scrapped the existing BI models and started over from source tables.
-
Fewer agents, more skills. The handoffs were killing us. Here's what the architecture looks like now and what we learned about the difference between "should" and "must."
-
I solved a distinct aggregation problem in Looker, documented it, and forgot about it. Months later the same constraint appeared in Cube. The solution was already in hand.
-
We migrated our embedded BI platform from Looker to Omni. I owned the architecture and the model layer. Here's what that actually involved: feature flagging, AI-assisted migration, and learning enough TypeScript to be useful.
