Recently, everyone has been talking about AI Agents running automated tasks on the chain, and I was tempted too, but the more I watch, the more I feel: if it’s really going to be implemented, humans still need to oversee it a few times. For example, during authorization, if you click the wrong one, it’s not “just trying,” but directly handing over the wallet’s private keys; also, on testnets, for the sake of points, repeatedly interacting, and when the script crashes, it starts clicking randomly, eventually creating a bunch of inexplicable historical records for yourself.

And then, strategies need to be monitored by humans; Agents can help you execute, but decisions like “whether to continue or when to stop” are hard for them to bear the consequences of. Especially now, with everyone guessing whether the mainnet will issue tokens, when emotions run high, machines are more likely than humans to blindly chase expectations. Anyway, my current approach is: automate where possible, but I must manually confirm permissions, capital limits, and exit conditions… otherwise, I’ll be reviewing the market until dawn, and all I’ll find are the pits I dug myself.