← field notes
Tag
#ai
Field notes on applied AI — agents and models wired into systems that matter.
3 notes
LOG-05RAG is not search with a nicer prompt
RAG reliability comes from retrieval: chunking, embeddings, permission checks, grounded answers with citations, and an evaluation loop.
aiarchitecture
LOG-04Letting an AI agent write to production safely
How to let AI agents write safely: idempotent writes, approval gates, blast-radius limits, reversibility, write budgets, and audit logs.
aiengineering
LOG-03When an AI agent uses your software
AI agents can now take real actions in your systems, not just answer questions. That's useful and risky. Here's how to let one work safely against a system that matters.
aiarchitectureengineering