There's definitely pieces of glue, boilerplate, OS-level application hooks, head project initialization that wouldn't be worth putting under test. And of course, other parts that you can trust to "just work". And then you have services and UI code that you might just want to stub out.

As for pure business logic, there would be a sweet spot somewhere around 80 percent, give or take, that is reasonable, depending on the project.

What most developers don't realize is that every if/else and switch case, break and continue statement in loops - they're the only ones that know about all of them. I doubt QA's test plan can get them all to execute through manual testing, if it's based on the requirements.

The important part of code coverage tools is that it can tell you in real-time which lines of code are covered not covered, and then you can decide if that's a problem or not in a case by-case basis.