These days, everyone is arguing intensely about whether the funding rate will flip or continue to bubble up, but I find myself thinking more about AI Agents: no matter how well they run, there are many steps that require human oversight. For example, during authorization, a single wrong signature is no longer just a "strategy mistake"; there are also cross-chain/bridges, contract upgrades, routing aggregation, on-chain slippage, and retry failures—Agents can automate these, but when anomalies occur, whether to keep executing or not needs someone to hold the reins. Then there's the information source: feeding it a bunch of data that "looks very much like the truth" might just lead to faster self-delusion. I'm tired but still here, so first, get the controllable permissions, limits, and whitelists in order, and then slowly observe which teams on L2 are truly taking security seriously.