ai-coding
- 22 days agoAI Coding
Near-Zero Hallucination Costs Half Your Answers
A RAG build that cut hallucination to 2% also stopped answering 54% of the questions it could have answered. The trade, the judge that enforces it, and what actually scales.
- 23 days agoAI Coding
Your RAG Leak Test Is Green and Your Access Control Is Broken
Nothing leaked. The permitted answer vanished instead, and every access test stayed green. Four ways RAG access control fails without raising one alarm.
- 25 days agoAI Coding
Four Ways Agent Context Breaks, and How to Detect Each One
Context failures look like a bad model. They are not. The four distinct failure modes, the symptom each produces, and the signal that tells them apart in production.
- 26 days agoAI Coding
Three Guards Every Agent Loop Needs, and How Each One Fails
The loop is the easy part. What a step limit, a validator and a confirmation gate each miss — found by an independent review, not by the tests I wrote myself.
