Recently, I've been looking at a bunch of AI Agents claiming they can help you automatically run on-chain operations. Honestly, I'm quite tempted, but the more I look, the more I feel that the "fallback" still needs to be handled by humans. For example, it can help you authorize and sign transactions, but who exactly is it granting permissions to, how much allowance is given, whether it's a one-shot unlimited approve—if it misses a detail like this, troubleshooting later becomes as complicated as solving a case. There’s also the contract call chain; sometimes it jumps an extra step or includes a strange route, and if the Agent just says "Completed," that’s too much of an oversimplification...

Recently, RWA and comparing on-chain yield products with US Treasury yields have also been quite popular. I tend to focus first on the flow of funds and whether the contracts are "clean." No matter how stable the yield looks, if the underlying assets, custody, or settlement chain can’t be clearly explained, I’d rather earn less. Ultimately, an Agent can run errands for you, but don’t rely on it to handle "authorization" and "proof"—don’t expect it to take the blame if something goes wrong. In the end, you still need to check the transaction records yourself. That’s all for now.