Recently I tried a few AI Agent–based protocols that automatically run interactions, and the efficiency is definitely higher—but honestly, some steps still need a human to cover for them. For example, if you fill in parameters incorrectly or the contract upgrade hasn’t been followed, the AI can just crash outright—then you need someone in the background who can manually cancel orders or pause operations. And about that re-staking “doll-within-a-doll” controversy—basically it’s like stacking blocks: each layer pulls away one block from underneath, and once you’re playing deeper into it, you have no idea which block is going to give way. When I reviewed audit reports myself, I saw those risk warnings—many of them are things AI can’t really judge, like “depends on administrator permissions”; you need humans to understand the context. Anyway, I’m the type of person who, no matter how fast the protocol is, still wants a manual switch. Learn as you go.