Picture a fraud model approving a card payment at 2:14 p.m. using a customer profile last updated at midnight. That’s how data warehouses, the backbone of business reporting, are built to work: they collect and update data in batches, often once a day. The schedule suited people reading reports, but it fails AI systems that make decisions in milliseconds, and Gartner predicts that 33% of enterprise software applications will include agentic AI, software that plans and acts on its own, by 2028, up from less than 1% in 2024. When Confluent, an IBM company, surveyed 4,000+ IT leaders for its 2025 Data Streaming Report, the biggest AI barriers involved data, not models: fragmented data ownership (68%), trouble adding new data sources (65%) and infrastructure that can’t process data in real time (61%).

IBM’s new whitepaper starts from that finding: every AI problem is a data problem. It explains how each wave of AI, from purpose-built machine learning to generative AI and autonomous agents, depends more heavily on fresh data. It then shows where a data streaming platform, which moves data between systems continuously as events happen, fits in the AI stack. If you own the AI budget, the data or the controls around it, read this before you buy another model.

Key lessons you’ll learn:

• The biggest blocker is organizational: 68% of IT leaders cite fragmented data ownership, meaning no single team controls data spread across many systems.
• Batch data limits AI: Warehouses that refresh once a day can’t support AI that acts in milliseconds.
• Agents are leaving the lab: 40% of organizations surveyed are piloting or deploying agentic AI. Those agents need to know what is happening now, not what happened yesterday.
• Fresh data reduces hallucinations: Large language models train on public data and lack your company’s context. Retrieval-augmented generation (RAG) fetches relevant company information before the model answers, and streaming keeps that information current.
• Governance must travel with the data: Apply quality rules, access controls and audit trails at every step as data moves between systems.
• Swappable parts limit lock-in: Treat models, agents and vector stores, the databases that help AI find related information, as modules you can replace when better tools arrive.
• The reported payoff: Respondents credited data streaming platforms with faster innovation (90%), faster time to market (84%) and higher revenue (84%). These are self-reported results from a vendor survey.