Recently, I've been seeing everyone talk about AI Agents automatically running strategies on the chain, but honestly, I remain quite calm about it. To put it simply, automation can save on operations, but there are still some parts that require human oversight: first, "who exactly is it interacting with?" If the contract address or routing changes, the Agent won't feel that something's off like a human would; then there's authorization and limits—I'm really hesitant to just give an unlimited approve automatically; and also, changes in pool structures (fee tiers, liquidity suddenly thinning out), impermanent loss—machines can calculate these quickly, but you still have to decide whether to bear the risk.

Moreover, the current on-chain data tools and tagging systems being criticized for being laggy or misleading isn't without reason. If an Agent relies too heavily on these "tag truths," it can easily follow false signals. My habit is: every time I need to add a new interaction target, I first try a small manual test, and at the same time, clear all authorizations... if it takes longer, so be it; anyway, I prefer to prioritize stability while seeking opportunities.