Recently, I've seen everyone talking about AI agents going on-chain to do work, claiming they can be fully automated. I’m not quite convinced... For real-world implementation, there are still several steps that require human oversight. For example, during the authorization step, no matter how smart the agent is, it might grant you an “unlimited limit,” and if the contract has a trap, it could be wiped out immediately; also, with cross-chain/exchange operations, routing issues, slippage, fees, and front-running can cause the machine to blindly push through with default parameters.

Then there's the incentive period for new L1/L2 projects to pull TVL. Veteran users complain that “mining, selling, and dumping” isn’t without reason: the agent might only recognize the highest reward pools, completely ignoring when liquidity is withdrawn or when others dump, ultimately turning you into a channel for exit liquidity... Anyway, my current approach is: the agent can help me run processes and monitor data, but I handle the final signing, limits, and withdrawal conditions myself. If it’s slow, so be it—I sleep soundly.