Articles

How AI systems actually behave

Short, practical pieces on the parts that decide whether a product works. Written for people who have to make one work, not watch it demo well.

Your RAG answer is wrong: retrieval or the model?

A three-minute test that tells you which half of a retrieval system failed, before anyone changes a prompt or swaps a model.

Read the article →

Most of what you call the AI is not the model

A map of the eight parts around the model, and a five-minute method for drawing it for any product.

Read the article →

Why a confident answer is not a verified one

How fluent text goes wrong, what confidence actually measures, and which outputs a product must check in code.

Read the article →

Autonomy is a cost, so decide where you spend it

How to choose between a fixed workflow, one model-chosen branch and a full agent loop, and where approval belongs.

Read the article →