Recently, I've seen everyone hyping up AI Agents that can fully automate on-chain work. I'm a bit tempted but also a little nervous... If I really let them handle on-chain interactions, there are at least a few steps I think still need human oversight: don't give out permissions carelessly (once you authorize/signature rights are given to too many, there's basically no turning back), and for contract addresses/routers, if the Agent makes a mistake once, it's not "being stupid," it's directly paying tuition. Then there's the extreme market conditions, like chain reactions in trading, especially when funding rates are twisted to absurd levels—an Agent might just mechanically execute strategies without understanding that the community's mood of "reversing or continuing to squeeze" can be quite deceptive. Anyway, for now, I mostly have it do monitoring and alerts; for real confirmations, I still do it myself... How about you?