Almost every day, articles appear warning that AI can only be successfully implemented if organizations have their data and metadata in order. Unfortunately, many authors fail to specify exactly what needs to be done. The crucial follow-up question “What does an AI-ready data architecture look like?” often remains unanswered. What architectural principles are needed? What metadata must be available? How do you make data understandable to both humans and AI agents? And how do you prevent AI solutions from getting bogged down in a collection of isolated experiments?
This session will answer these questions. Drawing on current developments in generative AI, agentic AI, and knowledge-driven architectures, we’ll discuss what a modern data architecture must look like to enable the deployment of AI. The focus here isn’t on the AI models themselves, but on the data architecture. From that foundation, governance, design, implementation, and management naturally fall into place.
- Why AI requires much more metadata than just simple definitions.
- The role of semantic metadata, business logic rules, and context.
- The automatic generation and enrichment of metadata using AI.
- Metadata as the foundation for RAG (Retrieval Augmented Generation).
- Deploying AI agents on source systems and the role of MCP (Model Context Protocol) within agentic AI.
- From metadata to knowledge graphs using data models, taxonomies, ontologies, and thesauri.
- Why traditional data lakes and lakehouses are insufficient as data architectures for AI.
- How AI can automate data lineage, impact analyses, and documentation.