Semantic-layer assertions require deep understanding of business rules and represent the current ceiling that AI-generated test cases struggle most to break through, which is also the fundamental reason why AI review cannot fully replace human review.