Lately I’ve been looking into AI agents and their interaction on-chain. Honestly, I’m pretty tempted, but I’m even more anxious… Things like automatic order placement, automatic currency swapping, and automatically running strategies can all be done, but when something really goes wrong, humans are still the ones who have to stand in. For example, I absolutely can’t bring myself to hand over the signing step completely. Even an agent that’s smart might still be “guided” by a phishing webpage into signing an authorization with an unlimited amount. Also, with cross-chain/bridges, contract upgrades, and permission changes—once you click the wrong thing, there’s basically no way back, and calling the police after the fact won’t help. Then there’s the kind of social mining and fan tokens where people “see opportunities and rush in.” Everyone calls it attention mining, but I always feel it’s more like emotional mining… an agent might be easier to get swept along by hype than people are. In any case, I’m more like someone who uses the drawdown curve as a sense of security, rather than someone who opens a position just because the timeline looks lively. First, keep the checkpoints that can be manually verified—sleeping well is better than anything.