Recently, watching AI agents interact with various on-chain activities, I feel they really work fast: scanning pools, comparing prices, breaking down routes, much busier than humans. But when it comes to the step of "whether to click confirm," it still requires human oversight—like giving permissions too broadly, not noticing contract upgrades, or signing content that looks like a jumble of characters... no matter how smart the agent is, it might treat you as if you’ve given a default consent button. Also, during extreme market conditions or network congestion, gas fees skyrocket, and strategies like whether to stop or withdraw liquidity are emotional choices that it’s hard for it to take responsibility for. Honestly, I now prefer to let the agent handle execution, while humans set boundaries: limits, whitelists, and position caps. Recently, with news of tax increases or compliance tightening and loosening, deposit and withdrawal expectations have also been fluctuating. In such times, I dare not hand over all the keys to autopilot—so for now, this is how it is.