Someone asked me whether AI agents can now operate on-chain, and whether humans are about to be laid off. I don’t think that’s the case. The three things that humans most need to cover for are still: First, “authorization/signature.” No matter how smart the agent is, it can still be led astray by prompts—handing it unlimited permissions is also awkward. Second, “liquidity exits,” especially when stablecoins slightly de-peg and the pools become shallow. It may only follow the procedure to swap, and then the slippage makes you question everything. Third, “exception handling”—contract upgrades, the front end going down, RPC acting up, and so on. Machines will only retry; humans need the courage to hit the pause button. Recently, developers have been quite excited about modularization and the DA layer, and it’s actually normal for ordinary users to be confused too… In short, the more layers there are, the more someone needs to take an extra look at key actions. I, for one, still don’t let it be fully automatic—I’ll treat it at most as a diligent assistant.