Artificial intelligence is rapidly migrating from laboratories to operations. According to a recent survey, by the end of 2025, half of companies used AI in at least three business functions. Copilots and predictive agents are being embedded in finance, HR, and logistics on an industrial scale, but a problem more important than GPU power has stood in the way of efficiency. As Irfan Khan, president and chief product officer of SAP Data & Analytics, noted, the modern AI crisis is a total lack of business context. Models work fast, but without an understanding of internal policies and processes, they lack judgment. The result is a high-speed conveyor of confident errors.
Old strategies with their giant "data lakes" have failed in the AI era: they methodically stripped meaning from information. For two decades, businesses dumped data into centralized repositories for reports, losing the semantic layer—the description of how the company actually works. To avoid feeding LLMs digital junk, management's focus is shifting from choosing the "smartest" model to creating a Data Fabric. According to Irfan Khan, this architecture doesn't just connect data across different applications and clouds, but preserves business logic. This creates a foundation on which AI agents can coordinate decisions rather than acting in a vacuum. For a logistics agent, this is the difference between a raw reaction to inventory levels and an understanding of risk policies or the nuances of working with suppliers.
Infrastructure, not the algorithm, now determines the return on investment. Launching autonomous agents into supply chains without a semantic layer is a threat to financial stability. Speed without judgment turns automation into a project that simply eats money. To avoid scaling chaos, managers will have to prioritize data architecture. As Khan summarizes, automation must reflect real business priorities, otherwise it will not help but actively hurt the organization.
Buying the most expensive AI model is a waste of budget if your data architecture remains a graveyard of context. By 2025, the divide will not be between those who have AI and those who do not. Competition will be between those who trust their agents and those trying to manage a fleet of hallucinating bots. If your data doesn't speak the language of your business, AI will never learn that language.