Snowflake VP of AI Baris Gultekin on Bringing AI to Data, Agent Design, Text-2-SQL, RAG & More

What It Covers

Baris Gultekin, Snowflake VP of AI, explains how enterprises deploy AI by bringing models to data rather than moving sensitive data to model providers, covering text-to-SQL breakthroughs, RAG implementation, agent design patterns, and predictions for autonomous knowledge workers in enterprise environments.

Key Questions Answered

• •Text-to-SQL Reliability: Reasoning models like Claude and Gemini now enable business users to query structured data directly without analyst intermediaries, achieving production-quality results on databases with thousands of tables and hundreds of thousands of columns by combining improved semantic modeling with enhanced reasoning capabilities that finally crossed the deployment threshold.
• •Unstructured Data Unlock: 80-90% of enterprise data exists in unstructured formats like PDFs and documents that were previously unusable. AI now extracts structure from these sources, enabling queries like finding contracts expiring soon in specific categories or analyzing quarterly results across ten years of documents through combined retrieval and analytics workflows.
• •Model Selection Framework: Frontier models like Claude handle one-off document processing, while Snowflake's specialized extraction models process hundreds of millions of documents at multiple orders of magnitude lower cost and higher throughput. Custom fine-tuned models only make sense when customers have unique data, strict cost requirements, and high-volume processing needs for tasks models haven't seen.
• •Data Governance by Design: Agents built on Snowflake automatically respect granular access controls, meaning the same sales assistant returns different results for different users based on their data permissions. This architecture eliminates data replication security risks while enabling broad deployment to 5,000+ users without creating new governance frameworks or security boundaries.
• •Product Development Transformation: AI coding assistants fundamentally change product management by enabling rapid skill prototyping instead of traditional UI development. Product managers now build working features in days, test with customers immediately, and only solidify consumer experiences after validation, inverting the traditional design-then-build workflow that dominated for twenty years.