I just looked at the on-chain operation records of a few AI Agents, and honestly automation is quite appealing, but when it comes to critical steps, humans still need to oversee. For example, authorization/limits—once you sign a large amount, it's like throwing the keys out the door; and with cross-chain or route switching, a change in slippage or path can easily lead the Agent to chase the "optimal" route and take you to strange pools. The most important things for humans to monitor are not clicking buttons, but understanding "why this transaction is being made, what's the maximum loss, and how to stop if something goes wrong." Also, now Layer 2s are arguing over TPS/fees/subsidies, which to me feels like a mirror of the market: they boast about performance on the surface, but behind the scenes, it's about who can keep people's attention... Anyway, I now only let the Agent handle small amounts, whitelisted contracts, and permissions that can be revoked at any time—just to keep things simple.