This is the way it seems to me too, but I just can't understand how a flaw this obvious could be overlooked by so many people? Surely we are both missing something?
In truth you can write code that is upgradable or ammendable, but always within limits of Ethereum transactions being immutable. However, when a project wants to emphasize that immutability, because that's perceived as the need by the users and the devs, then you end up in this situation.
So, as usual, the problem is solvable with a little diligence. The challenge is for crypto culture to get over itself and mature and actually perform that diligence.
I will say that there are very mature, very well developed projects that you don't hear about getting hacked, because they take advantage of the wealth of experience that's been built on this subject.
If you're going to potentially lose tens or hundreds of millions, you need a lot more than a little diligence. Formally proved code (something along the lines of Ada with Spark Pro) is the bare minimum for something with some much money on the line, and even then I'd still prefer a traditional contract and leave things to the courts.
However, if Ethereum sticks around there will be standardized, off the shelf contracts for people to use that have wide-spread testing/adoption. Once we reach that level of maturity then it could make sense to use it for things like escrow on a contract/purchase.
I can imagine using ETH for escrow; but there has to be a human in the loop for escrow, to direct escrow funds to be released. I guess one could contrive a situation where a "contract" could autonomously determine when it had been completed. But that's not the general case.
Escrow, or an oracle (checking, say, FedEx delivery.)
Most people use them for their projects, but you don't hear about them because they don't make it to hacker news as only crypto pessimism is upvoted here.