Recently I’ve been looking at IBC, message passing, and other cross-chain stuff. To put it plainly, “one-time cross-chain” isn’t really as simple as pressing a button: you don’t just have to trust the chain itself—you also have to trust the light client/validator set, whether the relayer is honestly running, the finality of the other chain, and whether the contract implementation is correct—because the “bridge” aspect is even more direct: adding an extra custodian or a multi-signature layer adds yet another layer of human checks and balances.

Now AI agents for automatic trading and automatic on-chain interactions are quite popular, and the narrative is flying high, but what I care about is exactly how they sign, how they limit permissions, and how failures and rollbacks are handled. Otherwise, if they do a bunch of operations and “automatically” drop both permissions and funds, I really wouldn’t be able to laugh.

My biggest fear isn’t slowness—at least if it’s slow, you can still see clearly. What’s scary is chaos: if it turns into a situation where a string of components keeps throwing the blame at each other, in the end you’re the one who has to bear it yourself. In any case, I only open perpetuals occasionally, and for cross-chain I’d rather do less—keeping the position curve uninterrupted is the most important.