For those of you running agents that chain several steps on their own, where do you put the human checkpoint: before the risky step, after the whole run, or nowhere because you trust the guardrails? Curious what made you land on that choice, because it seems like the call most people are quietly unsure about.