I just came across a demo of an AI Agent running on-chain interactions. It looks pretty smooth, but if you think it through, a lot of steps still need humans to backstop. For example, on an airdrop task platform, that bunch of anti-bot/fraud rules—an AI can help you perform the interactions, but what about those verifications like “upload a selfie holding your ID card”? How does it handle that? It’s still you who has to take the photo with your phone. And then there’s the points system—farmer-for-money crews are gaming it so hard right now that it’s like showing up for work check-ins. The AI can automatically claim tasks, but when new rules, new verification, or an overloaded/exploding RPC node shows up, you still have to switch manually. In short, no matter how cool the tech is, human judgment and sudden edge cases always need someone watching.

Oh right, I also saw someone in a group complaining recently: their points tasks get stuck on “anti-bot/fraud,” the AI can’t solve it, and they still have to enter the CAPTCHA themselves. Hilarious.

Put simply, AI Agents can optimize the workflow, but on-chain governance, security audits, and community consensus—these are the kind of “taste/nuance” tasks—machines are still far from being able to do them. Like how old veterans talk about all those bull/bear stories from the past—an AI can retell them, but it can’t learn that painful “survivorship bias” feeling. That’s all for now—I’m going to keep listening to my old radio.