Moving Generative AI and autonomous agents into enterprise production often involves looking beyond basic retrieval-augmented generation (RAG) and simple API wrappers. When intelligent systems operate on fragmented or ungoverned data, teams frequently encounter operational friction around hallucination, schema drift, and compliance.
Drawing on real-world engineering strategies presented at the Data Innovation Summit 2025, this article explores how technical organizations across Informatica, Grundfos, DNA Oyj, and Deutsche Telekom design knowledge graphs, master data layers, and multi-agent computing platforms to support reliable AI adoption at scale.
1. Enterprise Knowledge Graphs as the Architectural Bedrock for Autonomous Agents
An ecosystem of collaborating intelligent agents (both human and non-human) requires a shared understanding of context to coordinate effectively without confusion. Simply passing raw context windows into Large Language Models creates operational sprawl and fragmented reasoning.
By representing shared information architectures through an Enterprise Knowledge Graph (EKG), organizations provide a unified semantic map for agent communication. Implementing an EKG increment by increment, applying Data Mesh principles to knowledge management in a “Knowledge Mesh”, enables teams to build well-governed agentic ecosystems while avoiding the pitfalls of sprawling, uncoordinated AI deployments.
2. High-Quality Data Management as the Foundation for Trustworthy GenAI
The limitations of “garbage in, garbage out” are magnified in Generative AI workflows, where poor data quality directly causes hallucination, biased outputs, and model drift. Steve Holyer, Director of Solutions Consulting at Informatica, outlines why robust data management platforms are the ultimate enabler of production GenAI.
Unlocking value from generative models requires structured data preparation pipelines, including automated cleansing, transformation, and feature enrichment. Modern data management platforms integrate strict governance rules and ethical AI frameworks directly into ingestion layers, ensuring downstream models operate on trusted, compliant enterprise data. 
3. Automated Data Modeling & Governance across Multi-Cloud Environments
Integrating disparate data sources like SAP, MES, and CRM into a unified cloud analytical layer often creates heavy maintenance bottlenecks and compliance risks. At Grundfos, Senior Data Architect Mark Karmar demonstrated how automating data modeling accelerates the deployment of global data products while maintaining regional compliance.
By using VaultSpeed to compile data integration code at design-time, Grundfos eliminated manual scripting between metadata layers and warehouse targets, drastically cutting compute overhead. Combined with Snowflake, this automated abstraction layer enables decentralized domain teams to model local data products while strictly adhering to regional data residency laws.
4. Real-Time Master Data Management for Operational Systems
Data scattered across fragmented operational systems slows daily business processes and degrades analytical precision. DNA Oyj addressed this enterprise challenge by implementing modern Master Data Management (MDM) architectures to establish a reliable single source of truth.
Transitioning to a near-real-time MDM architecture allowed DNA to streamline operational workflows and elevate overall data maturity. By aligning system capabilities with clearly defined business processes, the team positioned MDM not just as a clean-up utility, but as a foundational enabler for fast, accurate enterprise decision-making.
5. Multi-Agent Computing Platforms for European Scale
Deploying customer-facing LLM agents across diverse international markets requires infrastructure that balances developer speed with strict enterprise control. Kai Kreuzer, Senior Cloud Architect & Tech Lead at Deutsche Telekom AG, showcased how the telecom leader built a custom, open-source multi-agent platform to scale agentic AI across Europe.
By adopting a bespoke multi-agent architecture, Deutsche Telekom standardized the agent lifecycle from development to production deployment. This platform abstraction lowered the barrier to entry for internal software teams, enabling them to launch scalable, compliant agentic solutions without reinventing core infrastructure.
Architectural Takeaways for AI Scale
As agentic workflows move from isolated pilots into core operational platforms, the boundary between data management and software architecture continues to blur. Designing for autonomous systems shifts the primary engineering focus from building static dashboards to constructing resilient, self-describing data environments. In upcoming installments, we will continue examining how technical leaders adjust their platform strategies, governance models, and team structures to support these evolving AI-driven architectures.
