Recently, everyone has been hyping up AI Agents that can fully automate on-chain actions, but I remain quite cautious. Frankly, it can help me monitor pools, scan announcements, and prepare transaction drafts, but when it comes to the final signature, authorization scope, and whether to chase that "seems very promising" opportunity, I still have to handle it myself. Especially with approvals—once given too much, if you don’t withdraw in time, your wallet gets drained first...

Lately, hardware wallets are almost out of stock, phishing links are rampant, and the more these happen, the more you shouldn’t leave the "clicking" to a robot. The biggest risk with Agents is mistaking fake official sites for real ones, or similar domain names for legitimate ones, and still thinking they’re helping you save effort.

A colleague also mentioned a couple of days ago that he almost clicked on an airdrop link in a group chat. After hearing that, I was even more convinced: no matter how smart the tool is, someone has to be responsible for "taking it slow." Anyway, my current approach is: automation only up to alerts and draft orders; before actually executing, I take a sip of coffee, glance at the candlestick chart, and review the authorization twice.