These days, I've been getting tempted again by AI agents, mainly because of that mindset of "It can help me click buttons now, maybe I can be a bit lazier"… but once it’s on the blockchain and running, there are still a few things that need human oversight, or else I might accidentally send myself away.

For example, during the authorization step, no matter how smart the agent is, it won't feel the limits for you, especially when dealing with new routes or new contracts. I definitely don’t dare give unlimited authorization; and for signing content, often the popup just shows a string of hexadecimal. Even if the agent says "it's safe," I still need to verify what exactly is going on myself. Then, when cross-chain or chain-switching, issues like Gas fees, slippage, and the state of bridges—if something goes wrong, it might just retry. I have to decide whether to "stop or continue" and whether to "change the route."

Recently, everyone’s been talking about social mining and fan tokens—this idea of "attention as mining." But I actually think: the first thing that gets mined away is my own attention… the agent can save on operations, but it can’t save your brain for monitoring risks. Anyway, my current approach is: let it run the process, but the key signatures and authorizations must be done by me. The rest, for now, stays like this.