In this talk, I present Intuit's approach to building AI-native customer experiences with a focus on interpretability and confidence. As organizations increasingly deploy AI systems at scale, ensuring reliable and trustworthy interactions becomes critical. I will discuss three key research directions that enable confident AI deployment: hallucination detection and mitigation, prompt optimization, and attribution, along with our proposed innovative work for addressing these challenges. These advances are already being implemented across Intuit's AI-powered products, serving millions of customers through platforms like TurboTax, QuickBooks, and Credit Karma. The talk will conclude with emerging research questions in building interpretable AI systems at scale.