Recently, I’ve watched AI Agents go on-chain to interact—it’s definitely hassle-free—but let’s be real: in many places, you still need a human to cover the gaps. For example, when it comes to the approval step, if the Agent says it wants to “increase the success rate,” it will just give you an infinite approve. I saw the allowance set to Max directly in the wallet pop-up, so I immediately changed it manually to a limit of 0.1 USDT—I'd rather sign multiple times. And for things like cross-chain transfers or contract upgrades, the Agent only follows the script; when the bridge gets stuck or the contract switches to a new implementation address, it doesn’t panic, but I sure do… In any case, my current rule is: automatic order placement is fine, but the steps of authorization, signing, and “sending the money out” must be watched by me. The whole setup in blockchain games—like inflation plus studio “brushes” that wreck the economy—also reminds me: no matter how smart the system is, when human nature and incentives get misaligned, the one left holding the bag in the end is still your own wallet. That’s it for now.