Recently, I've been hearing everyone talk about AI Agents automatically executing on-chain tasks. To be honest, I'm quite intrigued, but I still don't dare to fully entrust everything... Some steps definitely feel like they require human oversight: such as who gets the permissions, how to manage private keys/signatures, whether the contract will execute the same way during upgrades or parameter changes, and when the network is congested or MEV manipulations occur, whether to cancel the order or push through. These issues aren't something that a "single sentence model" can handle.

Thinking it over, the same applies to staking and shared security where yield stacking might get criticized as "layering" or "nesting." The more automated it is, the easier it becomes to silently accumulate risks.

My current approach is still low-frequency: stake when possible, use Agents as assistants, but I press the key buttons myself. Earning more slowly but sleeping peacefully.