Last night was a bit awkward... I asked a small agent to help me transfer tokens from the testnet to a new L2 address, but I accidentally missed two characters when copying the address, and my heart sank the moment I clicked confirm. Luckily, it failed on-chain and didn't go out, or else this "automation" would still be blamed on humans.

Honestly, now AI can help you check routes, calculate gas, and compare bridge fees, but I think there are still some steps that need human oversight: double-checking the recipient address, not giving permissions/authorizations unlimited access, and for irreversible steps like cross-chain or cross-L2 transfers, it's best to manually confirm twice. Especially when dealing with new contracts or upgraded proxy contracts, I prefer to be a bit slower.

Recently, everyone has been using ETF fund flows and US stock risk appetite to explain all the price movements. I also look at that, but it’s more like a mood thermometer; when it comes to actually making a move, the first thing is to get your own operational chain right, and not let "macro narratives" hide the most basic security details. Anyway, I’m currently doing small experiments, so even if something goes wrong, I can sleep peacefully.