I’ve seen this situation come up in almost every AI-assisted testing rollout. Natural language is powerful, but it’s also ambiguous, and sometimes KaneAI will misinterpret an instruction.
In practice, the system usually gives you two ways to handle it. First, you can rephrase or refine the instruction on your own, almost like teaching a junior tester who misunderstood the requirement. Over time, KaneAI tends to learn from these corrections within your project space, so it adapts without requiring a full retraining.