Am I the only one uncomfortable letting Claude directly call production APIs?

I've been spending a lot of time building examples with Claude Code recently, and one thing keeps bothering me. Claude is surprisingly effective at deciding what should happen. I'm a lot less comfortable letting it directly execute actions against production systems. A few scenarios that make me nervous: Accidentally posting duplicate comments because a tool call got retried. Generating a payload that's almost correct but violates an API schema in a subtle way. Calling an API with stale credentials. Trigger