Recently, everyone has been talking about AI Agents going on-chain to do automated work.

My first reaction isn't "cool," but "who's going to take the blame"...
Honestly, it's not hard for robots to press buttons; what's difficult is that they don't know when you're actually "scared."
For example, when swapping tokens on a DEX, if liquidity is thin and depth is shallow, slippage feels like suddenly stepping into a void at night.
They follow the strategy and still execute the trade, but your mindset can explode.
At this point, you still need someone to cover: how to set limit prices, how much slippage tolerance to give, whether to split orders or cancel, even the cowardly option of "not trading today."

And now, with the unlock calendar being refreshed daily, the selling pressure and anxiety spike, and Agents might just become more diligent in chasing volatility, ending up more like warming up MEV.
Permission management also needs human oversight—wallet authorization, limits, whitelists—don't let it give unlimited permissions just to save trouble, or you'll end up losing money and only be able to joke about it in poetry.
Anyway, my bottom line now is: automation is fine, but the risk switch must be in my hands; otherwise, the lessons from losses come too quickly to remember.