01AI cannot reliably use data it cannot access or whose meaning is unclear.

02Permissions, citations, freshness, tool access and human approval must be designed before model selection.

03A small measurable use case is a better start than a broad, uncontrolled chatbot project.