Recently, I’ve been seeing everyone hype AI Agents for automatically running strategies on-chain. My first reaction is still to look at the lending pools: actions like opening positions, topping up margin, and swapping collateral. If something goes wrong, someone has to stand behind it. Contract permissions being too broad means that if the agent glitches, it’s not “optimization” anymore—it could directly wipe out your positions…

Later, I thought about it and realized how ridiculous that is. I clearly wanted to save myself the hassle, but in the end I have to be even more on top of things.

And as for the “re-staking / shared-safety” yield-stacking setup being called out as a “Russian doll” scheme—I get it. The agent may only “see” one more layer of yield, but for edge cases like liquidity issues, liquidation thresholds, and oracle anomalies, you still need to set your own red lines: the maximum amount you can borrow, stop when a trigger is hit, and prefer making less profit rather than taking the risk.

Anyway, my current approach is to let it handle execution—I just keep the decision-making power and the emergency kill switch.