This argument sums up my recent thoughts on blockchains. Proofs of their trustlessness and eventual consistency rely on explicit assumptions (no cartels with >50% computation power, bounds on network partitions), but the underlying unstated assumption is that there is consensus about the protocol and its rules to begin with. We've seen this assumption break down with the block size debate and the DAO fork.

However, I don't think that this completely rules out their usefulness. Blockchains are more like "consensus abstractions" - if you have the underlying consensus (achieved through a governance model that can engender it), then it allows for uses like smart contracts that can efficiently work on top of it. The "trustless" property is also maintained - for example, even if a cryptocurrency is designed, issued and governed by a central bank, the fact that the network is decentralized and its integrity can be independently verified (even if it is centrally secured) still has some social benefit.