Just finished eating, scrolled through my feed, and saw a bunch of demos of AI agents doing automatic arbitrage and automatic interactions—talked up like it’s all just as real. Honestly, after doing MEV for so long, the thing I fear most is this “fully automated” narrative—how could there be so many perfect, closed loops on-chain?

For example, if you have an AI scan the mempool to front-run, the gas-bidding strategy alone is enough to make it pay dearly. And if it runs into sandwich robots that flip the tables, losses can come even faster than for humans. Then there’s the contract authorization step—many agents, to save on gas, directly grant infinite approvals, and when ordinary people don’t understand what they’re clicking, they end up getting drained.

Anyway, I’m pretty old-fashioned now: I only dare to use automated trading in scenarios with high determinism—like simple arbitrage or scheduled DCA. If I really need to deal with new protocols or do complex interactions, I’d rather go through the contract code myself twice, or use dedicated anti-sandwich tools as a backup. People’s biggest value is knowing when to back off. As for those who hype “fully intelligent” automation—eight times out of ten, they probably haven’t even been burned themselves. That’s it for now.