Lately, I've been seeing a bunch of AI agents claiming they can handle fully automated on-chain interactions, but I still feel a bit uneasy. When it comes to executing transactions/signatures, ultimately, humans are the ones who take the fall: details like authorization scope, limits, and validity periods—agents might "do" them, but not necessarily "do them carefully." My current habit is: for anything involving long-term assets or large amounts, I always manually review and authorize the target contracts, even if it’s slower.

And about the testnet incentives and point expectations—people in the group are guessing every day whether the mainnet will issue tokens… Agents can also help you run tasks, but when it comes to rule changes, black swan risk controls, or modifications to interaction paths, humans still need to keep an eye on things. Otherwise, a series of operations might end up being all for nothing. In the long run, I prefer to think of agents as "automatic homework-doing assistants," with humans responsible for the framework and boundaries—after all, habits are more reliable than talent.