Recently, everyone has been hyping up AI agents for fully automated on-chain surfing, but I’ve become a bit nervous… If it’s really going to be implemented, there still need to be some safety nets in place: first, permissions and limits—don’t just give unlimited approval right away; then, price sources/slippage—no matter how smart the agent is, it will make mistakes when liquidity is thin; and also, the chain reactions across protocols—if the collateralization ratio drops or liquidation is triggered, the contagion spreads faster than the agent can “react,” which is more deadly. In the group, screenshots of stablecoin regulation, reserve audits, and the back-and-forth of de-pegging rumors are circulating, plus the flood of rumors about de-pegging—when emotions run high, it’s easiest for the agent to execute commands that “shouldn’t have been executed” in the first place. To put it simply, I lower my expectations: let it handle repetitive tasks, while humans are responsible for braking and final confirmation—that’s a lot easier.