This Hacker News thread discusses the challenges of testing, particularly the common sentiment that "this can't be tested." Many argue that this phrase often translates to the cost of testing outweighing the benefits, especially in complex systems. There's debate on whether testability always equates to understandability, with some finding the opposite to be true. Specific examples like hardware drivers are discussed, highlighting the difficulty and expense of creating realistic mocks. Some suggest using hardware RTL simulations for testing, while others advocate for developers creating their own "fakes" for components. The discussion also touches upon snapshot/golden master testing, AI's impact on unit test creation, and the importance of end-to-end testing. It stresses that while AI can assist in test creation, the quality of the tests and focus on testing the right things remain critical. There's a divergence of opinions on the role of human testing versus machine testing, especially for user interfaces, but a consensus seems to form that they are both important, but each fills a different niche.