From the practice

Essays on how AI analytics actually behaves.

Written from the field: how governed systems get built, how ungoverned ones fail, and what the difference costs. Evergreen by intent — these are arguments, not news.

3 essays·Published when there's something worth saying. A shelf, not a feed.

Before the Copilot: What a Trusted Knowledge Layer Actually Is

Every AI analytics roadmap says 'governed knowledge layer' somewhere before the copilot ships. Here's what that layer physically is — the artifacts, the tests, the gate — in enough detail to build or buy honestly.

4 min read·knowledge-layer · semantic-contract · definitions-as-code

Your Analysts Are Becoming Human APIs

When definitions live in people instead of a governed layer, analysts become a request-response interface to their own data — high-latency, salary-priced, rate-limited by burnout. The fix isn't headcount.

4 min read·human-api-analysts · knowledge-layer · operations

Confident and Wrong: How AI Analytics Actually Fails

AI analytics doesn't fail loudly. It fails credibly — fluent, specific, wrong answers with no error message. Here's the failure mode, why better models can't fix it, and the layer that does.

4 min read·credible-failure · governance · evaluation

Also available as RSS.

Find out if your analytics layer is AI-ready.

Two paths: score yourself in five minutes, or bring the question to a working session.

Take the readiness scorecardRequest a working session