After returning from a night run, I checked the active addresses and bridge traffic on L2. The AI agent can indeed help me save quite a bit of time clicking buttons, but honestly, there are still a few things in on-chain interactions that must be handled by humans: the signature, the scope of authorization (I really don’t dare give it full permissions), and cross-chain or route-changing operations, which are hard to recover once made incorrectly. Recently, before and after the upgrade of that mainstream public chain, everyone has been speculating whether the ecosystem will move. I’m actually more concerned about whether the agent will continue to follow old strategies during “maintenance” or “chain reorganization” periods, because data delays can easily lead to overconfidence and mistakes.

What I regret isn’t the outcome, but that I was lazy and didn’t clearly define the risk control rules before letting it handle a few transactions. For now, let it be—automatic tasks should be automatic, but I still need to personally confirm key steps.