Long-form notes on production AI.

Architecture lessons, evaluation work, control vs. cloud, RAG/agents/evals - the work most teams skip until production proves they should not have.

• PRODUCTION AI2026-04-22

  Your demo worked. Now make it a system.

  The gap between PoC and production is rarely the model. It is data, governance, evals, cost, and latency - measured with intent.

  Read
• ARCHITECTURE2026-03-15

  Local is a decision, not a religion.

  When local beats cloud, when it does not, and the decision criteria that should drive the call.

  Read
• EVALUATION2026-02-10

  Evals first. Then everything else.

  Why golden sets, rubrics, and regression suites are the work that should have started on day one - and the cost of postponing them.

  Read