Lately, I’ve been seeing a bunch of AI agents talk about fully automated on-chain work—sounds great, but when it comes to lending, there are still a few places where someone has to stand by as a fallback. For example, if the price feed—or the oracle—has a small hiccup, the agent might keep ratcheting up leverage based on the “reasonable” data it’s fed. And if the interest rate curve suddenly turns, automatic rebalancing could push you to the side with the thinnest liquidity—then even withdrawing would mean lining up. And then there’s the step of authorization and signing: basically, once you hand over the keys, don’t expect it to always be able to clearly explain why it opened that door.

Not to mention something like attention rotation—today it’s memes, tomorrow a celebrity drops a couple of lines; the agent might chase hype faster than people, but your risk budget won’t magically grow on its own. My approach is still to check the protocol “weather chart” on the day—set the thresholds, limits, and circuit breakers ahead of time—then the agent can only run within the boundaries. If it goes outside the fence, it’s got to be my approval… Anyway, I wouldn’t hand it everything. What about you?