Just now, my phone popped up a red dot: some "AI agent has automatically completed on-chain operations," which looks quite convenient, but my first reaction was to check permissions and slippage... In plain terms, the agent can help you click buttons, but it can't necessarily take the blame. If something really goes wrong, those steps still need humans to cover: who did it give unlimited permissions to, whether the routing went through strange pools, whether it still executed trades when gas was wildly fluctuating, and most importantly — whose interests is it actually optimizing for.

Recently, with memes + celebrities shouting, attention shifts too quickly, so newcomers shouldn't expect the agent to help you dodge the "last mile"; it will only send you onto the train even faster... My current approach is a bit rough: large transactions must be manually confirmed, small ones let it run, but after each run I pull up a table to compare where the protocol's income/expenses went, at least to keep track mentally.