Recently, I’ve seen everyone talking up AI Agents running on-chain strategies automatically. To be honest, it saves some manual effort, but if I really let it run fully on its own, I’m still not quite comfortable. It does pretty well at querying pools, comparing prices, and splitting orders—but the moment it comes to “signing/authorization,” I feel like I need to cover it myself: whether the authorization scope is too broad, whether it’s granting unlimited approval, whether the contract address is correct… If this step goes wrong, there’s basically no way to recover. And for cross-chain, bridges, and cases where new protocols are launched with documentation that’s written in a foggy, hard-to-follow way, I’d rather move slower and click through it myself—at least then I know where the money is going.

Also, with miners/validators’ income, MEV, and ordering fairness getting criticized again, I feel even more strongly that an Agent can only help me run the workflow—it can’t take on the emotional and financial impact of getting grabbed/queued ahead of me. Things like slippage, gas limits, and retrying failed transactions are still boundary conditions that have to be decided by a person. Anyway, once I scaled down my goals, I ended up sticking with it longer: just let it do some mindless, brick-by-brick mechanical work, while I continue to be that annoying last “confirm” button. That’s it for now.