Recently, I’ve been watching AI Agents run on-chain interactions and, honestly, it’s pretty hassle-free—but let’s put it plainly: what it saves is “the act of clicking,” not “the responsibility.” The fallback steps still have to be handled by humans in my view: confirming the intent before signing (it says it’s switching to A, but the route actually detours to B), setting the authorization limits (don’t give an unlimited approve right away—otherwise you might wake up to a life-sized slippage/loss, like it’s disappeared in an instant), and making those little judgment calls like retrying on failure or adjusting gas. Agents can be a bit stubborn about it.

It also brings to mind that recent back-and-forth about NFT royalties… If an Agent only focuses on “the smoothest way to get a deal,” it might just treat creators’ income as friction costs to be smoothed away—liquidity feels great in the short term, but who’s going to sustain the ecosystem in the long run? In the end, it still takes people to make the call.

Personally, I trust data more, for a very simple reason: intuition in a bull market feels like talent, but in a bear market it feels like an illusion. Data at least helps me understand exactly where I went wrong.