Recently, everyone has been talking about AI Agents going on the blockchain, automated trading and such. The narrative is indeed easy to tell, but honestly, I'm more concerned about which parts still require human oversight. I think the signing and authorization steps should never be fully handed over to "automation," especially for unlimited permissions, cross-chain operations, and contract upgrades—when something goes wrong, it's basically your own fault for pressing the button. Then there are strategy parameters and stop-losses; no matter how smart an Agent is, it will be fed a bunch of "seems reasonable" data, and then it will step into pitfalls like slippage, MEV, blacklists, all on-chain. Right now, some people are busy touting that the future is already here, while others are focusing on security details. I, for one, prefer to take it slow... What do you think is the most important step to keep under human confirmation?