Recently, everyone has been talking about AI Agents helping you run errands on the blockchain.

I do want to save effort, but frankly, there are a few things that still need human oversight:
First is the signing step—no matter how smart the Agent is, don’t let it casually confirm transactions with hot wallets, especially for permissions you don’t understand or unlimited allowances.
Second is extreme situations, such as staking unlocks, unlocking calendars, or those days when they’re suddenly pulled out and scare you—on-chain slippage, gas, routing can all distort, and if the Agent rushes in according to rules, it might automatically suffer losses.
Third is review—an Agent will execute, but not necessarily reflect. I still prefer to treat each interaction as a pixel point to record, and if the noise is too high, reduce the position size or turn off automation.
That’s it for now. Automation is useful, but don’t outsource your thinking too.