How RAG fails in production, and how to fix it
The follow-up to the RAG explainer. The four ways RAG breaks in production (chunking, retrieval, faithfulness, a stale index), the fixes that hold, and the testing that shows them.
What is a context window, really?
A plain explainer of the context window: the shared token budget a model reads at once, why long context costs more and recalls less, and when retrieval beats a bigger window. About 12 minutes.
How to test an AI agent before you trust it
Test an AI agent in layers: rule-based tool-call checks, step scoring, a dataset from real work, a calibrated LLM judge, and expert review on the high-stakes cases. Five steps and a rubric.
Over 1,400 fake AI cases in court: where AI belongs
The same tool that does real work in contract review is getting lawyers sanctioned for citing cases that never existed. One habit separates the two, and it shows where AI fits in legal work.
Training AI on copyrighted work can be legal. Piracy isn't
The law on training data is settling into a rule. Training can be fair use, but how you got the material, and whether the output competes with it, can turn the same act into liability. Provenance is now the question.
What is agentic AI? A no-hype guide
Agentic AI plans and carries out multi-step tasks on its own. It decides what to do, uses tools, and keeps going until the goal is met. A clear guide to what really ships as an agent in 2026.
What is an embedding, really?
A from-scratch guide to what embeddings are, how they're compared, why they make modern search work, and which model to pick in 2026. About 15 minutes.
What is an LLM, really?
A from-scratch guide to how large language models work: tokens, attention, the prediction loop, and what to make of it all. 12 minutes.