My experience is mostly with server-side web frameworks, so that's how I'll describe it:

1. I start in an "outer" TDD loop:
2. Write a functional test that hits the URL of the feature I'm testing This fails for a silly reason: no route, no template, etc. (scaffolding)
3. I add only the piece of scaffolding that's causing the failure I'm seeing. At this point I'm just dealing with "glue" that the framework provides me.
4. Repeat steps 3 until I hit some logic that isn't solved with simply writing together things the framework gives me.
5. I call a function (or whatever) that doesn't exist yet but named for the logic I'm introducing. My integration test is now failing because it doesn't exist.
6. I start an "inner" TDD loop:
7. Write a unit test for that business function.
8. Make it pass as simply as I can.
9. Refactor/cleanup.
10. Repeat steps 7-9 until I'm satisfied that it's completely tested.
11. Pop back up to my outer loop. It should be passing now.
12. Refactor/cleanup.
13. Goto 1