Recently, I've seen a bunch of demos of AI Agents "automatically working" on the chain. Frankly, being able to run a process doesn't mean they can be responsible. Initiating transactions, signing, granting permissions—these steps, I still believe someone needs to have the final say: how much credit to give, how long to authorize, whether to stop when seeing suspicious contracts... These aren't problems that a "smarter" model can solve; it's about how much your wallet truly cares.

I'm not regretful about the outcome, but about the hassle I caused myself at the time by clicking through a bunch of unlimited approvals all the way. Looking back at the on-chain records, it’s like reviewing a medical record—more I look, more uneasy I feel.

Lately, everyone’s been arguing about rate cut expectations, the dollar index, whether risk assets rise and fall together or diverge. I feel that when macro sentiment gets heated, Agents are more likely to be influenced by your set "hype chasing" and go off course. My approach is pretty simple: I let it only handle information sorting and small-scale probing. When it comes to actually adding positions, switching pools, or cross-chain moves, I still do the confirmation myself—at least if something goes wrong, I can be the one to get blamed.