Watching a circle of AI Agents running on-chain interactions, it really feels like a "brain-saving but not worry-free" experience. Repetitive tasks like placing orders or swapping tokens are indeed faster, but whenever slippage occurs, routing worsens, or liquidity pools suddenly thin out, you still need to keep an eye on it. Otherwise, saving a few gas fees might cost you on the execution price. In short, the quality of execution isn't someone else's responsibility to cover for you.

And about authorization, Agent's one-click approve is very convenient, but if the limit is set too high, or if the contract is a phishing one, or if the signature content is wrong... these all require human oversight. Recently, hardware wallets are out of stock, phishing links are rampant, everyone's security awareness has improved, but the more "fully automated" it is, the easier it is to slip and accidentally confirm. Anyway, my current approach is: automation only handles small amounts and whitelisted addresses, and I review the key signatures myself—if it takes a bit longer, so be it.