"So near the physical boundary of the system there is a layer that requires fiddling. It is useless to try to write tests first (or tests at all) for this layer. The only way to get it right is to use human interaction; and once it's right there's no point in writing a test."
This seems dead wrong. There is probably no way to write tests first in this environment, but with so many different browsers interpreting your CSS (to run with preceding example) you need to be aware of when changes in your code cause changes in rendering that might need to be revalidated and further fiddled! I do agree that it doesn't fit well with TDD, but it absolutely can work with automated testing.