Blog

Shorter updates, lessons from client work, and opinions on shipping AI products.

• Office hours: what "production-ready" actually means

  Production-ready is not a feeling — it is a short, checkable list. Here is what actually belongs on it.

• A pattern I keep seeing in agentic AI pitches

  Capability is the easy 80%. The guardrails are the hard 20% that determine whether an agent is safe to deploy.

• Most RAG "evals" aren't evals

  A vibe check that happens to run once is not an eval pipeline. Here is what a real RAG evaluation process needs.

• The question that tells you if an AI feature is ready for real users

  A number beats a vibe. Here is the one question that separates a demo from a production AI feature.