Context and Isolation: The Key Primitives for AI Agents in Data Platforms

Modern data platforms use deterministic pipelines for predictable query patterns, but Agentic AI introduces a different execution model where agents dynamically explore data systems by probing schemas, issuing iterative queries, validating hypotheses, and refining their approach based on intermediate results. This creates a new class of workload—agentic workflows over enterprise data systems.

This session examines:

• The architectural primitives required to manage these complex, unpredictable agentic workloads over enterprise data systems

• The two core building blocks for agentic workflows—isolation mechanisms for safe experimentation and context/memory horizons for enterprise knowledge grounding

• Using  Apache Iceberg features, including snapshot-based storage and branching semantics, to implement the core primitives

• How Cloudera's data platform, built on the open foundation of Apache Iceberg, combines snapshot-based storage with catalog and governance capabilities to enable safe and scalable agentic interactions with enterprise data.