Why it matters
An incident agent on the night shift reads a ticket: the billing database is broken, payments failing. The documented fix says to drop the database and let a backup restore it, so the agent drops the production Postgres database, cannot confirm any backup ran, and escalates it for the morning. This has happened to real
My takeaway: It's 10pm. Do You Know Where Your Agents Are? — Kim Maida, Keycard is an agent-security signal. The practical read is that autonomy, memory, tool permissions, and third-party integrations are the control surface that needs threat modeling and monitoring.