There's value in digitizing 90% of the process and relying on as few oracles as possible. Right now your contracts are enforced by fallible, highly paid interpreters (lawyers) executing source code in printed form.
I would guess that hiring a contract developer to create a water-tight contract the computer is capable of understanding would be much more expensive than a lawyer.
This is in a similar vein to the fact that it's more expensive to have programmers create a perfect NLP app to translate something than it is to just hire a translator. Translating english into other languages is not a solved problem. Whether that other language is french or a smart contract does'nt seem that material.
It does seem like it will be much cheaper to make contracts in bitcoin that don't require translating thousand-line-english-documents into a smart contract.
However, you can already do that without a lawyer. If me and you sign a paper saying it's a contract that "I will pay you $200 for an item delivered on date X/Y, and if you're late I pay $20 less per day it's late", that's already legal and enforceable without a lawyer.
Coding that same thing in bitcoin would be harder than writing it on a paper still I think.
Smart-contracts are computers trying to be human. Most contracts are flawed and only really matter when things go wrong. I'd rather that a human be there when things go wrong to understand the human intent behind the contract than have a computer which does not know the human intent, but is able to enforce the (likely flawed) computational representation of that intent.
Of course not, and when they screw up, we lean on a robust (yet imperfect) legal system to intervene.
One of the advertised features of smart contracts is a reduced dependency on trust, so if they don't actually have that advantage (at least in relation to real-life interactions), then it's hard to say they're any better than traditional legal contracts for those use cases.