“LLMs are not reliable” Neither are humans, and we build systems around fallible humans to more reliably get the results we want: see routines and checklists for pilots and surgeons, construction inspection, accounting systems, audits, etc