Lately, I've been tired of seeing everyone interpret ETF fund flows, U.S. stock risk appetite, and crypto market fluctuations as if they're all tightly linked. Honestly, the more noise there is, the more suitable it is to use AI agents to handle mechanical tasks, but when it comes to on-chain interactions, humans still need to oversee a few key steps.

For example, during authorization and signing, even the smartest agent might mistake unlimited permissions for "efficiency." I now prefer to go slower, first verifying the contract address and permission scope, and isolating wallets if necessary. Then there's cross-chain swaps and slippage and route selection—agents can calculate these quite well, but when MEV and liquidity suddenly thin out, things can go wrong. In the end, you still need to set limit prices yourself and stop if it fails. Also, new protocols with frontends that look very similar—more phishing pages—make automation more prone to traps. Anyway, my current strategy is to let it handle bulk tasks when possible, I press the key buttons, and I take responsibility for stop-losses. That's it for now.