Recently, I've been looking into the AI Agent on the blockchain, and it's really cool, but after lowering my expectations, I feel more relaxed: it can help me monitor on-chain data, automatically split transactions, and even execute based on conditions, but in the end, the "scapegoat" is still the human.

The parts that most need human oversight: first is authorization/permissions, even the smartest Agent might give you unlimited authorization, and honestly, one mistake is enough to teach you a lesson; second is routing and slippage, those small MEV tricks on-chain might not be avoided every time; third is cross-chain/multi-chain operations, such as bridges, gas, retries on failure—if something really goes wrong, you still have to handle it yourself.

And now, between L2s, everyone is arguing over TPS, fees, and subsidies... Agents can also be tempted by "cheap" options, but cheap doesn’t mean stable. Anyway, my current strategy is: let it handle the chores, but I press the key buttons myself—just like that for now.