- Posted on
- Featured Image
A practical guide to replace ad‑hoc LLM chats with AI runbooks: versioned, auditable, Bash-first playbooks where an LLM proposes strict-JSON plans and commands, executed only with human approval. Includes a minimal YAML+Bash harness, Ollama/OpenAI setup, command whitelists, read-only defaults, examples (Nginx 502, K8s CrashLoop), and operational tips for testing, logging, and model lifecycle management.