Lately, I’ve been seeing everyone chat about AI Agents going on-chain to automatically get work done. I’m definitely tempted, but also pretty wary. To put it plainly, it’s not hard to automate the “click, click, click” part; what’s hard is who will take responsibility if something goes wrong: that authorization step, whether the contract address is correct, whether the routing has been stuffed with black-hole pools—those really need someone to do a final check with their own eyes, or else a whole string of signatures will just end up as tuition. And when the market shakes around, the Agent may stubbornly carry out the rules, while people end up hitting the brakes by instinct instead. On-chain data tools and the tagging system are also being criticized as lagging and potentially misleading, so I’m even less willing to let the machine make decisions based only on tags. Anyway, I’d rather go slower—leaving “manual confirmation” at the key steps. Even “when the dust settles, just be a block of stone” is fine too.