These past two days, the fee rates have been extremely erratic again—our group chat has been in an uproar: should we reverse, or keep squeezing the bubble… For my part, I’m still the same. I like placing single orders in spots where others wouldn’t dare to place them. But honestly, when AI Agents actually go on-chain and run, a lot of steps still need a human to stand by.

For example, it can automatically find pools, split orders, and even look like it’s pretty good at “timing.” But the moment the chain suddenly gets congested, gas spikes, or the router temporarily changes parameters, it’s easy for it to stubbornly crash straight through as if following a script.

Last time, I watched it route me through a weird path: the swap for 0x7b…c3f first jumped into a liquidity-thin pool that was basically unbelievable—the slippage blew up immediately. Luckily, I manually revoked the authorization and canceled the second leg of the trade, otherwise I would’ve paid tuition on the spot.

To put it bluntly, an Agent is suited to being a “diligent pair of hands,” but whether to force the move when it comes to risk-control thresholds, authorization scope, and extreme market conditions still ultimately needs a human to make the call. After all, when my orders are set to wait through a drawdown, I can stay calm—but if it gets excited, it definitely won’t spare the principal.