Lately, everyone has been hyping up AI Agents for fully automated on-chain work, but I feel a bit nostalgic: I've seen the scene where various "automated strategies" fail spectacularly... To be honest, most of the automation is just clicking a button; the real critical part still requires human oversight.

For example, the steps of authorization and signing—no matter how smart the Agent is, it’s hard for it to handle the consequences of "I’ve given unlimited authorization to someone"; and with cross-chain operations, route modifications, contract upgrades—no matter how well the documentation is written, ultimately, you still have to make the final call. Not to mention dealing with stuck transactions, front-running, or unfair ordering—retail traders complain that miners/validators are getting too greedy, MEV is too harsh, I get it—at such times, the Agent can at most help you cancel orders or reduce slippage, but you still need to keep an eye on things and not get carried away.

My definition of "long-term" is probably about a quarter; if it can survive a few narrative shifts without collapsing, then it’s somewhat promising... For now, just watch and see slowly.