I hate to admit it, but flaky tests almost always highlight weaknesses in my software architecture.

And fixing a flaky test usually involves making the actual code more robust.