Last night, I watched an AI agent automatically run workflows on-chain—it’s definitely convenient, but I still feel a little uneasy. I almost let a contract that “looks like an old address” through as a whitelist entry; only after I pulled the graph did I realize the funds had been routed two hops and ended up at a bunch of new accounts. Something felt wrong, so I immediately hit the brakes manually.

To put it plainly, the parts that are easiest to trick people with right now are still「authorization/limits」「which cross-chain route to choose」「what exactly the signature content is doing」—especially when a pop-up flashes by for a split second; if you’re not watching closely, it’s easy to just gloss over it. In the group lately, people have been repeatedly forwarding and keeping circulating talk about stablecoin regulation, reserve audits, and all kinds of “de-pegging” rumors. When emotions kick in, it’s even easier for the agent to go chasing hot-button operations…

What I do now is: let it automate as much as it can, but the very last step—the signature—I will definitely review myself. I’d rather move slowly than regret it. For now, that’s what I’m sticking with.