This morning, the traffic was a complete mess, and I only remembered to drink my coffee after it cooled down... It’s quite similar to these AI agents on the blockchain now—surface-level automation runs smoothly, but when you really hit a snag, you still need humans to cover the back.

I personally feel that the three areas most in need of human oversight are: first, “what exactly are they allowed to do,” because once permissions are expanded, a mistake during signing/authorization leaves no way back; second, “price and liquidity,” as agents jump at small slippages, but when the chain gets crowded or squeezed, the actual trade outcome is completely different from what they expected; third, “exception handling,” such as contract upgrades, RPC issues, or sudden routing failures—robots will just hit the wall according to rules, but humans need to hit pause in time.

Recently, watching the collapse analysis of blockchain games also gives a strong sense of déjà vu: inflation + studio exploiting the model to the max, coin prices spiraling, and all automated strategies amplifying the volatility. Anyway, I now believe more in “taking it slow + being able to stop at any time,” rather than outsourcing the safety net to a program.