Do you still read your agent’s approval prompts, or do you just hit yes?

I thought about posting this bc yesterday I was finishing the newsletter for the new version of an app, with an agent helping. It pulled the changelog, wrote the copy, fixed the links and asked me to approve each step along the way. Somewhere around the 30th prompt I was just hitting yes on everything. One of those yeses wasn’t “send a test to myself.” It was “send to the list.” So everyone on it got a release email with a link to a changelog page that didn’t exist yet. The agent did exactly what I approved, I just didn’t read it... And that's the thing with clicking “allow” on every command. It gets old fast, and after a while you stop reading, which kind of defeats the point of being asked. Full access isn’t the answer either. It’s fine until the one time it isn’t. The setups I’ve seen people use: (1) An allowlist, so it only asks about commands it hasn’t seen before. (2) Asking only when a command looks dangerous. (3) No prompts at all, but the agent lives in a container. (4) A read-only plan first, then you approve the whole plan instead of each step Each one trades a bit of safety for fewer interruptions, and I can’t decide which trade I’d rather make. Curious to find out what your setup is, and has skimming through a prompt and incorrectly approving ever failed you like it did for me?