Session

Sponsored by: IBM | Extend your Databricks Architecture with Real-time Context

Overview

ExperienceIn Person
TrackData Engineering & Streaming
IndustryConsulting & Services
TechnologiesAI/BI, Delta Sharing, Unity Catalog
Skill LevelIntermediate
DOWNLOAD SESSION SLIDES

AI agents can dramatically accelerate your ability to react to new information and make trusted decisions — but only when they have the right context from your data. However, preparing data sets for each individual AI use case is time consuming and expensive — hindered by fragmented data landscapes, inconsistent governance, and data complexity. In this session, learn how to power AI agents with real-time context from unstructured and structured data across hybrid data landscapes — without migration. Empower your AI agents, and your business, to act with confidence.

Session Speakers

Ed Anuff

/CPO, Data, IBM Software
IBM

Erik Duffield

/Partner - Americas Data Leader
IBM (HQ)

Full Summary

Extending Databricks with real-time, governed, hybrid data across the enterprise

Enterprises often discover that agentic AI underperforms not because of models, but because the underlying data is stale, fragmented, or stripped of governance and meaning. The conversation argues that teams already invested in the Databricks Platform can unlock more value by pairing it with real-time streaming, zero-copy hybrid access, and governance that runs at execution time across cloud, on-premises, and mainframe environments. For teams evaluating how agentic and analytical workloads fit together, enterprise AI agent trends, top use cases, governance, and evaluations illustrates how leading organizations are structuring that investment.

FAQ


It builds on, rather than replaces, current lakehouse work. Zero-copy formats, federated access via watsonx.data, and real-time streaming with Confluent feed governed data into the Databricks Platform so existing pipelines and agent development continue while the reachable estate expands to mainframe, on-prem, and sovereign sources.

Streaming carries semantic metadata and enforces policies inline. That turns the stream into a real-time context engine and a runtime governance point, giving agents the meaning and controls they need, not just event delivery.

Human-curated catalogs cannot keep up with derived and agent-generated data. Policies must be machine-readable, semantic-aware, and enforced at runtime so agents cannot route around literal rules.

Many signals and references live in documents and other unstructured assets. Agentic RAG patterns, including IBM's OpenRAG on OpenSearch, complement structured federation so agents can reason over both within a single governed architecture.

Yes. Mainframes and other on-prem systems continue to generate the freshest, most business-critical data. Bringing that data into the agent's world model without forced migrations is key to accurate, enterprise-wide decisioning.