1: Snowflake for Gen AI Overview

Learning Objectives

By the end of this chapter, you will be able to:

• Map a generative AI use case to the correct Cortex surface across Cortex Models and Functions, Cortex Fine-tuning, Cortex Search, Cortex Analyst, and Cortex Agents.
• Choose between Cortex Code in Snowsight and the Cortex Code CLI for development workflows including Cortex Agents and Snowflake Intelligence orchestration.
• Decide when Snowflake Copilot Inline accelerates author-time productivity and how cross-region inference and the CORTEX_ENABLED_CROSS_REGION parameter affect latency and availability.
• Position Snowflake Intelligence as the turnkey natural-language interface and integrate it through REST APIs and the Model Context Protocol.
• Select between AI Studio, SQL, and the REST API as Cortex interfaces, and apply Cortex Knowledge Extensions and Cortex Code CLI commands appropriately.
• Bring custom and open-source models into Snowflake using the Snowflake Model Registry and Snowpark Container Services.
• Reason about prompting, vector embeddings, and context windows when calling Cortex AI functions.
• Apply Cortex Search across multi-index retrieval, access control requirements, and the supported invocation paths.

Executive Summary

• Snowflake's Gen AI architecture is a stack of managed surfaces (Cortex Models, Cortex Search, Cortex Analyst, Cortex Agents, Snowflake Intelligence, Cortex Code) and BYO surfaces (Snowflake Model Registry, Snowpark Container Services) that all run inside the customer's governance boundary.
• Each Cortex surface exposes a distinct contract: AISQL functions are per-call generative endpoints, Cortex Search is an indexed retrieval service, Cortex Analyst is a semantic-model-grounded text-to-SQL endpoint, and Cortex Agents orchestrate the three plus Snowflake Tools.
• Cortex Code and Snowflake Copilot Inline target developer-time productivity; Snowflake Intelligence targets end-user natural-language analytics; cross-region inference is the toggle that decides where the model serves.
• Bring Your Own Model paths split cleanly: the Snowflake Model Registry tracks artifact lineage and inference contract metadata, while Snowpark Container Services hosts the runtime for open-source models, fine-tunes, and vector sidecars.

Assumptions

• The reader has hands-on experience with SQL, Snowpark Python, RBAC, and warehouse sizing on Snowflake.
• All examples assume an Enterprise edition account in a region where Cortex AI is generally available; preview features are flagged inline.
• Region and edition caveats matter: features in Public Preview behave like GA from an API standpoint but should not be used for production SLAs without verifying the release notes.
• Throughout the chapter, function names with the SNOWFLAKE.CORTEX. prefix and AI_ prefix refer to the same family of AI functions exposed under both namespaces.