> If you want a reliable system you should know when to fail

I think the main problem is that most people pushing for heavy AI delegation either

1. Don’t really care about system reliability or 2. Don’t understand the difference between code and system, and assume that a system must be reliable if some automated checks pass