Yesterday I was messing around with an AI agent to help me handle on-chain interactions—it’s convenient, but when it comes to the crucial steps, you still need a person to cover for it. For example, with authorization, it only ever says “need approve,” but who exactly is being approved, what the allowance amount is, and whether it’s an unlimited approval or not are things I have to watch myself. Otherwise, one slip and my wallet turns into a “shared bike.” And for cross-chain/bridges, the agent sees low fees and wants to go, but it can’t tell me about the bridge’s past incidents, or whether the contract was just updated. Not to mention airdrop tasks—rules get changed today and again tomorrow. The agent runs according to the old process, and only after it finishes do you find out it was all for nothing. Right now, developers are pretty excited about modularization and the DA layer, and it’s totally normal that ordinary users are left confused too… Anyway, I just remember this: automation is fine, but don’t get lazy with signatures and approvals. For now, that’s it.