AI Agents can do work on-chain by themselves now—so should we humans be laid off and go clean cat litter? I think it’s still too early… even the cat that steps on my keyboard knows that some pots the machine can’t lift.

To put it simply, Agents are great at “execution,” but not so great at “responsibility”: how much you authorize (approve) for, whether the contract is a phishing scheme, whether the routing detours into some weird pool, and whether the cross-chain bridge is truly safe—once something goes wrong, it’s not something a simple retry can fix. And when it comes to slippage, failed retries, and Gas spiking uncontrollably, someone has to hit the brakes.

Lately, everyone’s been complaining that validators/miners are getting too well-fed, and that MEV and ordering are unfair—I feel it too. No matter how smart an Agent is, once it gets pulled into the black box of ordering, it can only take the hits. So my habit now is: let it run small amounts, in batches, on whitelisted contracts, and for the key steps I still confirm myself… I’ll start by posting a cat picture to calm down first.