Your point is valid. In real-world work, tests should focus on parts that are difficult to verify, and if everything passes on the first try, it's often a sign that something deserves a closer look.

That said, what I wanted to highlight in the example was a contrast — tools like Cursor and other general-purpose models often fail to even generate simple tests correctly, or can't produce tests that pass. So the goal was to show the difference in reliability.

Related to this, consider that when an LLM writes tests for code, it's writing them based on what the code actually does, not what it's supposed to do. This is equally true when the code itself was written by the LLM. Sure the tests pass, but that doesn't prove the code is correct.