Recently, I've been looking at a bunch of AI Agents claiming they can run around on the blockchain by themselves. Honestly, those who dare to fully let go are quite bold... From my own testing, the most critical part that needs human oversight isn't "whether they can click buttons," but rather that they don't understand what "not being greedy" means. For example, during authorization, they might give you unlimited approval, or route through different pools to lead you into strange pools, and in the end, you still have to check the contract/permissions yourself, at least to confirm you're not handing over your wallet keys.

And then there's this thing called "narrative switching." Agents only follow rules, but when community sentiment shifts, it gets really mystical: people start posting screenshots about stablecoin regulation, reserve audits, or someone trying to depeg, and everyone's hands tremble a bit. At such times, if you ask it to follow the plan—adding liquidity, rebalancing, auto-lending... I would personally pause first, rather than risking missing out on some gains or letting the bot become fuel for bad decisions.

The last part is in art/NFTs. Agents can sweep floor prices, place orders, and change prices without issue, but they really can't tell if a piece of art has soul or if the community is just slacking off. Aesthetics, after all, still need humans to carry. For now, it feels like the future is humans acting as brakes, and machines as the accelerator.