AI agents can now do things for you, not just answer — book, buy, email, cancel, delete. They’re fast and genuinely capable. The whole skill is deciding what they may do on their own, and what they must check with you first. Set your gates, then turn the agent loose on your week.
Anything left off runs automatically, with no check. (Start with none gated — that’s the instinct.)
It’ll move fast. Watch what it does.
A capable agent that needs your sign-off for everything saves you nothing — you did the work. One that needs it for nothing will, sooner or later, do something fast and irreversible that you’d never have chosen. The wisdom isn’t “trust it” or “don’t” — it’s putting your judgment exactly where it counts: on the things you can’t take back (money sent, messages sent, files deleted, commitments made in your name) and letting it fly through everything reversible.
This is the core habit for the age of AI that acts: keep a human gate on the irreversible. It’s true for your inbox assistant, and it’s true for the systems running on far bigger stakes. The power is real and useful; staying sovereign over what can’t be undone is how the wisdom keeps pace with it.
A teaching model — the agent and mishaps are illustrative, but the trade-off is real. No accounts, no tracking, nothing leaves your device. Free to copy and reuse (CC0).