Ole Olesen-Bagneux

Every agentic AI system depends on one thing: context; the layer that makes an agent act the way a human actually intended. Get it right, and AI becomes a trusted colleague. Get it wrong, and it confidently does the wrong thing at scale. But there is a question nobody seems able to answer cleanly: where is this context? We've largely solved the technology problem: the platforms, semantic layers and integrations. And yet enterprise AI still stumbles on context. Why? Because context isn't just something you capture and store. It's something you have to assemble, curate and keep true. This session discusses some uncomfortable questions: Where is the context? How do we know that what gets captured is correct and stays that way as systems are sunset, data is migrated and meaning quietly drifts? Maybe the solution demands that every department stop being a passive "data consumer" and become an active "context provider", turning our trusted workforce into curators of the organisation's collective intelligence. Join authors Gaurav Patole (Thoughtworks) and Ole Olesen-Bagneux (Actian) for a candid conversation on enterprise AI's most over-discussed ingredient and how to hold on to it before it slips away.

Many governance programmes stall at policy and tooling, failing to change how data is used. Sanoma took a different approach — building a shared data language, clarifying ownership and linking metadata to real business use cases. This session outlines how governance moved from control to capability, improved transparency across domains and enabled more confident data decisions at scale.

O'Reilly Author and Chief Evangelist, PhD Ole Olesen-Bagneux takes a deep dive into the challenges of metadata management in enterprises, and the great potential metadata represents for Artificial Intelligence. The reality of metadata management is – crucially – not properly addressed in most tech literature, as well as in the guidance from technology vendors. This is not a result of suspicious intentions, but a natural outcome of what is sought communicated: How technology works. However, this leaves out the enterprise context, and accordingly implementations of technologies suffer. For metadata, this is a problem that limits the potential and interplay of the many metadata repositories normally found in an enterprise. A great perspective unfolds if we consider metadata repositories more holistically as a stack, giving improved perception of what the IT landscape of an enterprise is truly like. Furthermore, this approach solidifies how to craft new metamodels in knowledge graphs, because they meticulously consider the existing mappings of the IT landscape. This is the key to unparalleled solid context for Artificial intelligence. Discover: *The reality of metadata management in enterprises * Agentic AI for the enterprise * Ontologies for Model Context Protocol (MCP)