Over the past couple of days, everyone’s been talking about AI agents handling on-chain interactions. It feels pretty cool, but also quite “deadly”: letting it automatically rotate pools, claim airdrops, and perform borrowing and lending really is convenient—but when it comes to that moment where you have to sign, I still want to cover it myself. Especially things like authorization limits, whether the contract is newly deployed, and those transactions where a bunch of calls are stacked together. To put it plainly: one mistake isn’t just “missing out on a bit of profit”—it’s a direct loss.

The testnet incentives and points campaign makes that even more obvious. Agents are the most “diligent,” but whether the mainnet will actually distribute tokens, and whether the rules will change—no one can guarantee that. It can help you run through the process, but it can’t tell you whether it’s worth spending time and taking the risk. Anyway, right now I only let it do information gathering and simulated execution. Before moving any positions, I go through the permissions and routes manually again. If it’s slower, then it’s slower—let’s do it like this for now.