I use test doubles (mocks, fakes, stubs, whatever) and I still distrust them. Test doubles let you run your core logic quickly against your collaborators based

I use test doubles (mocks, fakes, stubs, whatever) and I still distrust them. Test doubles let you run your core logic quickly against your collaborators based on assumptions about how they behave. You also need integration tests to validate that those assumptions aren't wrong. It's not either/or, it's both, mindfully applied.