By Stephanie Simone
As AI agents move into real-world enterprise use, the challenge is no longer just building models, it is building the data foundation that allows those systems to understand context, retain memory, and act.
DBTA recently held a roundtable webinar, Data Foundations for Agentic AI: Enabling Context, Memory, and Action, with Kelly Kohlleffel, senior director, global partner sales engineering at Fivetran and Jeff Mills, senior product marketing manager at Redis, who discussed how organizations are designing data architectures specifically for agentic AI.
These architectures include:
- Semantic layersto deliver consistent, trusted business context
- Knowledge graphsto connect data and enable deeper reasoning
- Vector and retrieval systemsto unlock unstructured data for AI
- Modern data platformsto power scalable, AI-driven workloads
- Real-time data pipelines and metadata systemsto enable continuous, action-oriented AI
According to Kohlleffel, data now powers AI agents, operations, and real-time systems, not just reporting. However, the modern data stack cannot support continuous, scalable, AI workloads. The forces behind this shift include platform convergence, AI-driven change, and scale and complexity.
The data foundation for AI needs to consist of:
- Build for AI agents: Fresh, complete, governed data. Ready for agents to act on.
- Always on: Spend time acting on data, not waiting for it.
- Open by design: Your data. Any engine. No lock-in.
- Engineered to evolve: The right foundation today won’t be the rebuild you regret tomorrow.
Move, manage, and transform data with Fivetran, Kohlleffel said. The Fivetran platform can integrate scheduling coordinate transformations with data movement; provide SQL-based transformations that run directly in the destination; quickstart data models speed analytics; give operational visibility into transformation runs, and more.
The pressure to implement agents is stifling, Mills said. And modern data architectures can fail at runtime, completely breaking context in four ways:
- Incomplete answers get escalated
- Selling sold out products lose money
- Slow answers equal churned users
- Forgetfulness breaks trust
Mills suggested Redis Iris, a real-time context engine that offers a Context Retriever that makes data navigable. Users can store history and preferences and retrieve key details for agents and customers. It can also keep context in Redis synced from other data sources and improve agent results with accelerated retrieval using fewer tokens.
For the full webinar, featuring a more in-depth discussion, Q&A, and more, you can view an archived version of the webinar here.