Recently, while browsing for tea, I came across a bunch of AI Agents on the chain, and it felt pretty cool, but when it comes to actually spending money, humans still have to take the responsibility. For example, the signature step: it can help you calculate the path, compare prices, even automatically split orders, but contract permissions, authorization limits, whether it’s leading you into a strange router… if something really goes wrong, it’s just a one-click return to the pre-crypto era. And about cross-chain/bridges, no matter how smart the Agent is, it can’t stop you from clicking the wrong network, entering a phishing frontend, or encountering sudden network congestion causing slippage to spike. The incentives and points on testnets are even more obvious; everyone’s guessing whether the mainnet will issue tokens. The Agent just follows the tasks, so you need to think carefully whether you’re doing an interaction or just gambling on expectations. Anyway, I mostly have it observe and draft orders now; if I really need to execute, I’ll take another look at authorization and deposit/withdrawal paths… that’s all for now.