Enterprise AI has rapidly evolved from predictive models answering specific questions to generative models producing content on demand, and now to agentic AI systems that autonomously pursue goals through multi-step reasoning. Each phase of AI development requires different data infrastructure approaches, with predictive AI focusing on high availability and low latency, generative AI on blending pre-training with context through systems like retrieval-augmented generation, and agentic AI demanding flexible, responsive, and trustworthy infrastructures to handle increased complexity and real-time data needs. Companies like PayPal and Wayfair have successfully leveraged predictive AI for fraud detection and personalized shopping experiences, respectively, while Myntra uses generative AI to enhance customer interactions with its conversational shopping assistant, Maya. As agentic AI introduces greater infrastructure dependency, organizations are urged to design systems that ensure data discoverability, session-scoped caching, and real-time data freshness to support this advanced AI stage effectively.