Recently, everyone has been talking about AI agents going on-chain to automatically do tasks, and I’m quite tempted, but I still don’t dare to fully let go. Checking if a contract is a scam, how broad the authorization scope is, which cross-chain bridge to use, and then signing that final step... I think these steps must be overseen by a human, or else you might step into a trap without even knowing how. Honestly, agents can help me “run errands,” but they can’t take the blame for mistakes.

I used to be quite stubborn, always saying I only look at on-chain data, but I found that’s not enough: on-chain data may look healthy, but if community sentiment collapses, liquidity can disappear just like that. Now I see it the other way around—focusing only on sentiment makes it easier to be manipulated. The recent debate over NFT royalties is a perfect example: some say they need to protect creators’ income, but others complain that secondary sales lack liquidity... In the end, the ones who suffer are ordinary users caught in the tug-of-war between taking on risk and creators losing revenue. Anyway, I prefer to take it slow—give fewer permissions if possible, do things manually when I can, and start with that.