Recently, I've seen everyone hyping up AI Agents that can fully automate on-chain arbitrage and other tasks, but I actually want to ask: which parts still require human oversight? Honestly, I still don't dare to entrust signing to it, especially with pop-ups full of incomprehensible permissions, unlimited quotas, and such. No matter how smart AI is, it can't take the blame if something goes wrong. Also, for routing, cross-chain transactions, and contract upgrades—when encountering "seems similar" fake addresses or phishing sites—an Agent might just follow the process without stopping, but a person needs to pause and verify. And recently, there's been a lot of heated debate about privacy coins and mixing services; the boundaries of compliance are quite blurry. Automating interactions with such services through an Agent would be even more complicated... Anyway, my current approach is: it helps me monitor data and spot opportunities, but I make the final click myself—if it takes a bit longer, so be it.