Recently, a bunch of people have been hyping up AI Agents that can work on the blockchain by themselves, but honestly, just clicking a button doesn’t mean they can take responsibility. If I were to believe in them, at least these steps still need human oversight: the signature (it doesn’t understand how dangerous “unlimited authorization” is), cross-chain/exchange (slippage + routing issues can cause chaos), and the verification before contract interaction—“who exactly are you clicking on?” Especially now, with memes + celebrities shouting buy signals one after another, newcomers are most likely to let the agent chase hot trends until the last moment, and the fees are paid upfront.

I personally see “signals” as very crude: whether the authorization list has inexplicable extra items, whether the wallet pop-up’s recipient/contract address matches, whether there’s a sudden string of failed transactions on the blockchain. Anyway, an agent can serve as a quick-handed assistant, but not as a driver. If something goes wrong, it won’t send you a tx hash to apologize.