AI Agents can now handle on-chain actions by themselves, but honestly, I still don’t dare to fully let go. The step of checking the contract address still requires human oversight; otherwise, it might authorize just because the “name looks similar,” and that’s how phishing scams happen—changing the avatar to trick users. Plus, there are authorization limits, slippage, cross-chain routing—Agents might pick the cheapest-looking path, but end up stuck on a strange bridge for a long time without progress. And don’t even mention the signature step—I’d rather be a bit slower and review what I’m about to sign; if I make a mistake, it’s just tuition… but I don’t want to pay that tuition.

Additionally, the modularization and the DAO layer narratives are exciting developers a lot right now. Users (including me) just want to ask one question: after I click “Confirm,” who is actually holding my data, and who do I contact if something goes wrong? It’s fine if Agents can automate the process, but responsibility boundaries, exception handling, and the “stop” button still need human oversight. A slower response is actually better—at least it prevents me from impulsively handing over my wallet.