Why Your Data Architecture Cannot Support Agentic Analytics

The Semantic Layer: Why Business Context Is Critical for Agentic Analytics

Raw database schemas are insufficient for AI consumption. Column names and table structures lack business context.

AI agents need to understand what data means in your business context. Without this, agents guess at definitions, misinterpret relationships, and deliver answers that look right but are wrong. A semantic layer defines business concepts once and maps them to underlying tables across all systems.

Effective semantic layers are: integrated into the query engine for programmatic access; store business logic in one place; enable agent-accessible queries through MCP or SQL; and enforce row-level security ensuring agents only see authorized data.

Dremio's AI Semantic Layer provides this foundation. Business context is defined once, enforced everywhere, and accessible to both humans and AI agents. Built into the query engine, not bolted on.

PL Developments: Building the Architecture for AI

PL Developments faced the same challenge as many organizations: their legacy architecture could not support AI. Manufacturing 5,000 OTC pharmaceutical products for major retailers, they had hundreds of ETL jobs consuming 30% of ERP resources. When they wanted to deploy AI agents for customer communications and quality processes, the fragmented data landscape made it impossible.

They adopted Dremio as a unified data access layer, connecting operational data in Postgres with analytical datasets in their Iceberg lakehouse in real time. With this foundation in place, they integrated Cursor AI directly via MCP (Model Context Protocol) Server, giving AI agents secure, governed access to unified data with full business context.

Results: PL Developments deployed LLM applications that generate customer communications in under a minute and query FDA databases for vendor warning letters in real time. By retiring legacy ETL, they freed 30% of ERP resources and reduced complex analytical queries to 6 seconds.

The 2026 State of the Data Lakehouse & AI

65% of data leaders prioritize agentic analytics for 2026, but 70% say siloed data and weak governance are blocking progress.

According to Dremio's 2026 State of the Data Lakehouse & AI Report, 92% plan to move most analytic and AI workloads to the lakehouse within the next year. Nearly half cite lack of unified, AI-ready data. 40% highlight poor data quality and missing semantic definitions. As organizations deploy agents, shared definitions and metrics are no longer optional but essential for reliable AI operations.

Rethinking Data Architecture for Agentic Analytics

Over the past quarter, I have met with dozens of Chief Data Officers. I ask every data leader: "If you were designing your data architecture today knowing AI agents would generate unpredictable queries across all your data sources, would you make the same architectural decisions?"

The consistent answer is no.

Organizations effectively deploying agentic analytics have converged on similar principles: query federation across sources, semantic layer integrated into query execution, access controls enforced at query time, autonomous query optimization, and agent-native interfaces including MCP support. A media company CDO told me: "We federate queries in real-time. This eliminated pipeline delays and reduced data warehouse costs by 40%."

"The question is not whether to modernize your data architecture. The question is whether you will do it before or after your competitors do."

Sendur Sellakumar

CEO, Dremio