Recently, I saw AI Agents running on-chain "doing their own work," honestly kind of like new interns: they can do the job, but the mess still has to be borne by humans. Automated signing, dollar-cost averaging, arbitrage—those are fine, but when it comes to changing parameters, switching strategies, handling exceptions (like a contract throwing a strange error, or unclear authorization scopes), I still have to manually confirm, or else I wake up the next day to find my wallet turned into "public property," which would be awkward.

And as for governance, Agents can help me scrape forums, summarize proposals, even simulate voting results, but I’d rather click the final vote myself, after all, disagreements aren’t something you can just calculate—they’re something you argue out and then compromise… I still believe that "human intuition is sometimes more reliable than models."

By the way, a quick rant about modularization and the narrative of the Data Access layer—developers get excited just talking about it, but for users it’s basically: as long as you’re happy. My current bottom-line principle is: before money leaves, someone must understand where it’s going. That’s all for now.