Just now, my phone popped up a red dot: a certain Agent said "automatically completed on-chain operation"… My first reaction wasn't peace of mind, but rather to check what exactly it signed.

Now AI is quite smooth handling "clicking buttons/running processes," but it's still early for me to fully let go: authorization limits, which route to take, slippage and retry on failure, as well as dealing with abnormal contracts/phishing links—ultimately, someone still needs to verify and confirm.
Especially recently, everyone is staring at the unlock calendar every day, and the pressure to unlock causes anxiety to spike. The Agent might mechanically execute according to strategy, but it could just happen to hit the most illiquid times.
Anyway, my current approach is quite crude: manually go through key steps, give permissions sparingly, split into batches when possible, to avoid waking up and discovering that "automation" has taught me a lesson.