In the past few days, I've seen a bunch of AI Agents automatically demonstrating on-chain interactions. To be honest, I'm quite tempted, but I don't dare to let it run completely automatically. The part that most needs human oversight isn't "whether it can click the button," but rather: who is it interacting with, how much gas is spent, whether it will repeatedly retry and burn through the transaction fee if it fails, and that signing step—once signed, there's no way to revoke even if you regret it.

Especially when new L1/L2s are trying to incentivize and boost TVL, on-chain congestion is like rush hour. Agents might only look at "can the transaction go through" and not consider "is this transaction worth that much cost." In the end, they become mere support tools for mining, arbitrage, and selling. Right now, I see myself more as someone calculating with a small notebook rather than handing my wallet over to a script and going with the flow. It's fine to have it as a reminder or order placement assistant, but if I really let it run wild, there should at least be a whitelist, limit caps, and manual secondary confirmation. Otherwise, waking up to find a gas bill that’s more expensive than my meal money is just too much.