Recently, everyone has been hyping up AI Agents for fully automated on-chain operations, and I’m tempted too, but honestly, someone still has to take responsibility. The most critical part is authorization—when the Agent sees “saving effort,” it often grants unlimited permissions, and then you realize your wallet is sleeping with the door wide open; there are also cross-chain/swapping pools, where a change in route can cause slippage to skyrocket, and if no one’s watching closely, it’s a disaster. Not to mention contract upgrades, blacklists, fake frontends—Agents can also fall for phishing links.

I’m actually a bit slow on MEV and fair ordering… A couple of days ago, I was still wondering why the transaction prices always seemed weird, and then I realized I got sandwiched again. I’m not very good at calculating miner/validator revenue either; my gut feeling is that “the money I pay is being eaten up happily.” So my current fallback method is pretty crude: I manually approve large transactions, and let the Agent handle small ones; I only give a one-time limit each time, preferring a bit more hassle over becoming Exit Liquidity again.