Recently, the hype around AI agents interacting on the blockchain has been quite lively, but honestly, the real implementation still depends on humans to take responsibility. For example, with signatures—authorization scope, validity period, whether it can be reused—no matter how smart the agent is, it’s easy for it to just follow the process and then send itself away; and then there are sneaky things like routing, slippage, and MEV—when transaction density on the chain changes, bots often become more stubborn than humans. Right now, I mostly let it help me monitor depth and calculate order placement ranges; when it comes to actually executing, I still press confirm myself. By the way, recently, before and after the upgrade of that mainstream public chain, everyone in the group has been guessing whether projects will migrate or not. I think it’s better to see who’s front end or node can’t hold up first during the maintenance downtime… I no longer believe that “full automation custody equals peace of mind,” because in the end, the one who always takes the blame is that string of private keys in the wallet.