Recently, a bunch of people are hyping up AI Agents automatically running on the blockchain, acting like full custody asset management... I definitely don’t dare to try. If I were to use it, it would only be for “monitoring the market + drafting”: for example, first extract the contracts to interact with, authorized addresses, call data for me to review, and the final signature must be manually confirmed by me, especially for approve/permit transactions, where a careless mistake could grant unlimited permissions. Also, links, RPC, pop-up signature instructions—no matter how smart the Agent is, it can’t prevent phishing websites that change domains. Honestly, what can be automated is the process; what’s not can’t be automated is responsibility. If something goes wrong, the blockchain won’t recognize “AI clicked for me.” Let’s see.