As enterprises move from AI experimentation to intelligent applications and agents, context engineering is emerging as a critical foundation for success.
Delivering the right context at the right time across structured data, unstructured content, business rules, and real-time signals is becoming essential for improving AI accuracy, relevance, and reliability.
DBTA recently explored this in a recent roundtable webinar, Context-Driven AI: Building Unified Data Foundations for Smarter Apps, with experts who discussed strategies for connecting fragmented data environments, enriching AI systems with business context, and enabling smarter, more responsive applications and workflows.
Robert Stanley, senior director of special projects at Melissa Informatics, said the core challenge is AI is accelerating, but confidence in the context and data behind it isn’t. “Mostly confident” is not a safe foundation for context-driven AI.
According to Stanley, key context elements for AI apps include the following:
- Application requirements — documented roadmap, logic, training resources.
- Business expertise — rules, constraints, MCP, domain knowledge.
- Data environment — access, quality, currency / recency (freshness).
Safe and effective AI can deploy workflows and interpret results, but it cannot replace competitive insight into application requirements, he said. AI does not replace validated expertise. It amplifies whatever business logic you give it, whether good or bad.
Small data issues can become large AI failures, Stanley said. Dynamic data quality is required for effective AI.
Stanley recommended teams create a well-defined application grounded in human expertise and intuition; document, protect, and publish proprietary business logic and resources; and invest in dynamic, automated DQ/MDQ.
Without data unification, AI lacks trusted core data, said Suchen Chodankar, principal product manager at Reltio.
Without enough context, Silos and misaligned policies fragment what apps can see, agents face a higher risk of poor decisions, and teams get pulled back in to recover.
Chodankar explained that with a governed task view it provides one identity, relationships, and recency; rules, permissions, and evidence; and the smallest trustworthy context for the task. Entity resolution creates the first layer of trusted context.
Reltio’s AI-powered matching engine (FERN) finds difficult candidates. Rules, thresholds, and reviews keep high-risk merges safe.
Jeff Vestal, senior principal AI architect at Elastic, noted that AI just made vector search a must-have for everything. Every AI application needs to retrieve context. Every agent needs to query data. The vector database is the new database.
Data and context is siloed across the enterprise. Elastic bridges the gap between enterprise data and high-quality AI experiences. Elastic Inference Service provides low-latency GPU inference at scale. Users can build AI features with zero setup, have model choice and orchestration, and provides the shortest path to search, RAG, and agents.
Elastic offers a continuously optimizing knowledge layer between your data and any agent, Vestal said.
Ebrahim Alareqi, principal machine learning engineer at incorta, said the model is rarely the bottleneck. What context requires is live, detailed, and connected data. If you miss any one of the three and everything built on top of it inherits the gap, he said.
The hardest data to reach is the data AI needs most. Structured operational data sits inside the ERP, behind schemas built for running transactions, not for reading them. Unstructured and streaming sources are comparatively easy to bring in. This is the piece that decides whether any of it can be combined at all.
For the full webinar, featuring a more in-depth discussion, Q&A, and more, you can view an archived version of the webinar here.