Just now I saw someone in the group say again, “Cross-chain is just a click and it’s there”… To be blunt, every time I do cross-chain, the first thing that goes through my head is: who exactly am I trusting? With something like IBC—relatively “rules-based,” at least it transmits proofs and the results of consensus. What you trust are the light clients/validator sets on both chains, and don’t mess around with relays (in theory they can’t fabricate, but they can still delay). If you switch to many bridges, it turns into trusting multi-signatures, trusting oracles, trusting a whole set of external verification networks—the more components there are, the more places things can go wrong.

Recently, this wave of AI Agents automatically interacting with on-chain has been pretty noisy too. Some people only know how to hype “automated, hands-off arbitrage,” but I’d rather ask this: for the cross-chain/message-passing module that the Agent calls, where exactly is the trust boundary drawn? Anyway, I’d rather go a bit slower now, take a closer look at the paths and verification methods, and don’t let “automation” also package the risks for you.