Recently, I’ve seen AI Agents go on-chain to run interactions, and it feels pretty convenient—but when the moment comes to actually take money out of my own pocket, someone still has to back it up. For example, permissions and limits—basically, don’t let it open up the whole wallet in one go. Then there’s the signing step. The Agent might see “Approve” and think that’s normal, but I need to take another look at who it’s approving, how much, and whether it’s granting unlimited authorization. And with things like cross-chain and swapping pools, if the route changes, slippage can make people laugh. The machine may pick the “optimal” option, but when it encounters congestion or MEV phishing, it also may not know to “do nothing for now.” Last night, a colleague even complained about new L2 incentives to pull up TVL—ultimately it turns into a one-stop “mine/withdraw/sell” flow. The Agent is diligent, but I actually want to move slower; I’d rather make a little less than give up my sense of security. That’s all for now—I’ll keep doing low-frequency dollar-cost averaging. For other interactions, if possible, I’ll do them manually.