Recently, I've been seeing a bunch of AI agents claiming to automatically do on-chain arbitrage, governance voting, and the like. Honestly, when it really runs, someone still has to cover the bottom line. For example, in the signing/authorization step, no matter how smart the agent is, it might "conveniently" give you an unlimited quota. If something goes wrong, you won't even know who to blame; and with slippage, routing, cross-chain bridges—when congestion and abnormal rollbacks happen, machines tend to execute rigidly according to scripts, which is quite mechanical and costly.

Then there's the whole sorting and MEV stuff. Recently, retail investors have been complaining that validators/miners are making too much money, which I can understand. Agents just follow the mempool and compete, and when they get front-run or sandwich attacked, the experience isn't much different from manual confirmation, sometimes even worse. My bottom line right now is: let it do "monitoring + alerts," but I make the key transactions myself, at least to keep my mental account from being drained by automation.

First, I’ll revoke the authorizations for a few old dApps, so I won’t forget them again someday.