Good piece and thoughts:

> Accountability becomes diffuse. When a system fails and no human reviewed any step, who is responsible?

(1) My current belief is that the human holds the final responsibility - it was the human who decided to trust (whether there is verification happening or not). When I set Claude Code on a task and focus on the functional requirements instead of the implementation, it was still my choice.

(2) On the scenarios, I would expect for the 5th one to dominate over the others - based on anecdotal evidence at work, and as it looks like the most logical option.