Smart contracts are being suggested as the building block of "autonomous corporations" (implying "do anything!") yet nearly anything meaningful basically has to interact with the real world through agents external to the contract for most business cases. Given this severe limitation, could someone clarify what the whole shebang is supposed to be about?

I actually saw Vitalik speak last week or so here in Shenzhen came away less convinced than ever of their theoretical utility for anything but a tiny fraction of users.

In short, adding a decentralized network of paid/incentivized actors to any existing potential cryptographic problem space (a fair rough summary of the smart contract notion?), in particular the subset relevant for most businesses, doesn't seem to solve anything particularly well and typically decreases critical measures of engineering elegance such as simplicity, comprehensibility, predictability, etc. while increasing negative measures such as technical lock-in to piles of rapidly evolving technologies for which hiring and building is expensive and error-prone.