Recently, watching AI Agents run on-chain interactions, it’s indeed more convenient, but frankly, many steps still require human oversight. For example, during authorization, the Agent might give unlimited permissions for the sake of “smoothness,” and if you don’t watch it, it could set a trap; also, with cross-chain routing, gas fees, slippage, and which pool to use, it looks optimized, but when abnormal market conditions hit, it’s easy to go off course. Not to mention the signing content—humans can at least glance at the domain name and amount, but sometimes the Agent is too “confident,” and even if it’s wrong, it still signs.

These days, everyone’s talking about rate cut expectations, the US dollar index swinging along with risk assets. I definitely don’t dare to hand over all decision-making; macro changes can amplify those small probability on-chain incidents.

I’ll first review the authorization history of my commonly used wallets and change unlimited permissions to limited ones.