Recently watched a demo of AI agents on-chain interaction, it’s indeed convenient, but honestly, it’s still a bit early for full automation. For example, during authorization, who exactly is the signature given to, and how much quota can I spend? No matter how smooth the agent explains, I still have to double-check myself, or else one day I might not even realize my permissions have been nested and extended. Also, with routing/slippage, the blockchain can change its face instantly; the path the agent chooses looks optimal, but in reality, it might get squeezed, so ultimately, a human still needs to set a bottom line.

What’s more annoying is “exception handling”: RPC timeouts, contract upgrades, cross-chain stalls… Most agents just retry, and eventually, it turns into emotional management. My approach is more conservative—keep smaller positions, let it operate within familiar protocols, so if something deviates, I can stop it in time.

By the way, recently there’s been a lot of noise about social mining and fan tokens—this “attention as mining” concept. It seems more like packaging human impulses into KPIs. Agents can handle some execution, but questions like “Should I do it, is it worth it, do I trust it”—those still really need humans to backstop for now.