I spend my days building agentic systems that actually ship — not demos, not prototypes. Real systems, in production, with real constraints.
Twenty years across the full data stack: pipelines, warehouses, lakes, semantic layers, and now the agent runtimes that sit on top of all of it. I've seen what breaks at scale and what quietly just works.
The hardest part of AI isn't the model — it's the data trust, the rollback plan, and the 3am alert. This blog is where I think out loud about what's working and what's not.
Strategy by me. Research assisted by LLMs. Humans stay in the loop — always.