I’m not very good at chasing hot trends everywhere, but lately I’ve noticed that AI agents are getting more and more capable of clicking around on-chain, and I still feel there are a few steps that must have a human backing things up. For example, when it comes to granting permissions, even if the agent is smart, it might still get you to sign an “unlimited allowance”—plainly speaking, that’s handing the keys to someone else. And for things like cross-chain transfers and switching pools, with slippage, routing, and contracts—whether it’s still the same old playbook—machines may not necessarily care about your fees the way you would. Plus, when you run into anomalies (getting stuck, being sandwiched, or suddenly being prompted to upgrade), it’s more important to stop than to push through.

Airdrop season now feels like clocking in for work again. Task platforms are getting tougher on anti-sybil measures more and more—agent automation is convenient, but once the rules change or your account gets flagged, in the end you still have to decide whether to keep grinding… These days, I’m more inclined to let it handle “execution,” while I handle the “braking” and risk control. Besides, if we slow-cook it a bit, I can sleep more soundly.