Someone asked me, since AI agents can now run on-chain by themselves, what do humans still need to do as a safety net? Honestly, the main focus for humans should still be “before placing an order” and “after something goes wrong.” Before placing an order, don’t let it hold unlimited permissions to sign everywhere, set limits, whitelists, and simulate transaction paths (especially when dealing with sneaky tactics like front-running). These really need human confirmation at a glance; after something goes wrong, don’t expect it to calmly cut losses on its own—when stuck in pending, gas prices spike wildly, or routing suddenly changes, quick human intervention to cancel orders or switch channels is still faster. As for the “automatic execution according to rules” part, just let it do its thing, but you need to clearly define the rules.

By the way, regarding the recent NFT royalty debate… agents can certainly help you automatically place orders and sweep floors, but the value judgments—like whether to support creator royalties or prioritize liquidity—are not something they can argue with you about. Ultimately, it’s up to you to choose which side to stand on. Anyway, my current approach is: let the machine handle the chores, humans are responsible for hitting the brakes.