Recently, I've seen everyone talking about AI Agents automatically on-chain operations, and it seems quite convenient, but I still don't dare to fully entrust it. The signing step must be backed by a human, otherwise, they could set the authorization to unlimited, and even casually authorize a stranger's contract—then you'll be crying before you know it. Also, details like cross-chain transfers and parameter modifications that are "almost wrong chain"—no matter how smart the machine is, it will mistake "similar" for "correct." The staking unlocks, token unlock calendar, and other issues have been brought up again, which is quite alarming. I found that Agents can also be influenced by this kind of emotion; the more fear of selling pressure, the more frequently they want to rebalance. Later, I thought it was quite funny—anyway, I still stick to my old habit: only let it do reminders or simulations, and I confirm the actual order myself to sleep more peacefully.