I think the point about knowing when not to answer is one of the most important parts of building trust in AI. Customers can usually forgive an AI for saying, “I’m not sure, let me get someone who can help,” but they’re much less likely to forgive a confident answer that turns out to be wrong.