My phone just popped up a red dot again: an “AI agent has automatically completed on-chain operations.” It looks pretty convenient, but my first reaction is to go check permissions and slippage… To put it simply, an agent can help you click buttons, but it may not be the one who takes the blame. If something really goes wrong, those steps still need humans to cover: who it granted unlimited permissions to, whether it routed through those odd pools, whether it forced trades when gas went haywire, and—most importantly—who’s interests it’s optimizing for “the best” (i.e., at the optimum).

Recently, with memes plus celebrities making a quick shout, attention shifts too fast—newcomers shouldn’t expect the agent to help you dodge the “final baton.” It’ll just get you onto the train even faster… My approach is a bit old-fashioned: for big amounts, you must confirm manually; for small amounts, let it run. But after every run, I pull up a table to check where the protocol’s income and fees went, so at least I’m clear in my mind.