Recently, a bunch of people have been claiming that AI Agents are fully automated on the blockchain. Honestly, they dare to let it run on its own, mostly because they haven't been taught by MEV. Automation can handle tasks like placing orders or adjusting routing, but when it comes to signing authorization, setting limits, whether the contract is a shell swap, or hiding malicious parameters, humans still need to oversee it. Otherwise, one mistaken authorization could mean permanent exploitation. Also, when the chain suddenly gets congested or prices jump wildly, the Agent will just follow the rules and push through, ending up caught in a sandwich. Then you still have to manually stop losses or revoke permissions.

By the way, I want to complain about those recent big on-chain transfers and exchange hot/cold wallet movements being called "smart money coming in." I think it's more that people are just eager to find a story... Who owns the address, how the path is taken, whether the transactions are split, or if it's an internal fund collection—ignoring these details and jumping to conclusions is like fortune-telling.

I personally trust data more. Not because data is always correct, but because it at least allows for replay analysis; if your intuition fails, you won't even know where you started to drift. Anyway, Agents can serve as co-pilots, but not as fully autonomous driving.