Recently, I’ve seen everyone chatting about AI agents running around automatically on-chain—sure, it’s convenient and saves effort—but I still feel like some parts have to be handled by people as a fallback. For example: the signature step—how much permission are you granting, whether the contract address is a phishing trap, whether the cross-chain bridge is acting up again today… If you leave all of it up to “its own judgment,” and something goes wrong, then you can only blame yourself for being too lazy. Also, when it steps in during trading slippage or when liquidity is thin, the loss isn’t on the model—it’s my sleep.

I also thought about that whole NFT royalty water war argument. To put it plainly, it’s a clash between “automation rules” and “human favors/accounting”: if you want liquidity, someone wants to turn off royalties; if you want to nurture creators, then someone has to be willing to pay that extra. Anyway, my approach right now is pretty crude—let it automate everything it can, but for the key step, I still click to confirm myself. Even if it’s slower, it feels reassuring.