Agentic AI is Snowflake’s answer to a very practical problem: “How do we let people ask for outcomes in natural language, and have AI do the work safely on governed data inside the Data Cloud?”
Snowflake’s Agentic AI Framework brings together four pieces:
Snowflake Intelligence – the agentic front-end and control plane
Cortex Analyst – natural-language analytics on structured data
Cortex Search – retrieval over documents and other unstructured content
Cortex Agents – the orchestration layer that plans, routes, and executes tasks
Together, they let you move from “build a dashboard” to “ask an agent” – without sending data out of Snowflake.
1. What Snowflake Means by “Snowflake Intelligence”
Snowflake Intelligence is the user-facing experience (and account-level object) where agents live and where business users interact with them. It lets users:
Create charts and get instant answers in natural language
Work across structured and unstructured data in a single experience
Use agents that are connected to semantic views, semantic models, Cortex Search services and other tools (Snowflake Documentation)
All queries run under the user’s Snowflake role and warehouse, so existing RBAC and data-masking policies continue to apply automatically.(Snowflake Documentation)
From a business point of view, Snowflake Intelligence is the portal where agentic AI becomes visible: a curated list of agents that different teams can use for insights, exploration and operational workflows.
2. Cortex Analyst – Text-to-SQL, But Enterprise-Grade
What it is:
Cortex Analyst is a fully managed, LLM-powered service that provides a conversational interface to your structured data in Snowflake. Users ask questions in natural language and get answers without writing SQL. Behind the scenes, it converts text to SQL with a focus on reliability and high accuracy. (Snowflake Documentation)
Why it’s “agentic”:
Snowflake describes Cortex Analyst as part of an agentic setup: it uses LLMs plus semantic models to reason about the user’s request and decide which fields, joins and filters to apply. (Snowflake)
Key characteristics:
Semantic models in YAML bridge business terms (“store”, “policy”, “channel”) to the underlying schema. (Snowflake)
Multi-turn conversations let users refine or follow up without restating context. (Snowflake)
A classification step can reject or clarify ambiguous questions instead of hallucinating a confident but wrong answer. (Keyrus)
Typical use cases:
Natural-language BI for executives and business analysts
“Explain this number” or “why did X change?” queries
Lightweight self-service analytics where a full dashboard is overkill
3. Cortex Search – Retrieval for Documents and Knowledge
Where Cortex Analyst focuses on tables and views, Cortex Search focuses on text and documents.
Snowflake positions Cortex Search as a fully managed search service that combines:
Vector search for semantic similarity
Keyword search for lexical matching
Reranking to push the most relevant results to the top (YouTube)
This is the engine you use when you build RAG-style applications or enterprise search over content stored in Snowflake (manuals, PDFs, knowledge bases, FAQs, product docs, etc.). (YouTube)
Typical use cases:
“Ask the policy manual” assistants in insurance or banking
Support agents searching across troubleshooting guides
Internal knowledge portals for HR, legal, or operations
4. Cortex Agents – The Orchestrators
Cortex Agents are where the “framework” becomes truly agentic.
According to Snowflake’s docs, agents:
Parse a user request and plan a solution, breaking it into subtasks
Route across tools, using Cortex Analyst for structured data and Cortex Search for unstructured sources
Execute the plan, aggregate results, and generate a final answer
Iterate and monitor, with detailed traces for auditing and debugging (Snowflake Documentation)
In other words, instead of a single prompt-in / answer-out, agents can:
Understand what the user is really asking
Decide which data and tools are needed
Call multiple tools in sequence
Return a well-reasoned answer, not just a snippet of text (Hakkoda)
This is what enables scenarios like:
“Compare last quarter’s performance, explain the drivers, and link me to the underlying policy changes”
“Summarise key risks from these claims and show supporting stats”
5. How the Pieces Fit Together
The user interacts via Snowflake Intelligence – typically in a browser at ai.snowflake.com, selecting an agent appropriate to their role or domain. (snowflake.help)
The chosen agent (Cortex Agents) plans the task – identifies whether it needs structured data, documents, or both, and decomposes the request. (Snowflake Documentation)
It calls Cortex Analyst when it needs reliable text-to-SQL over semantic models. (Snowflake)
It calls Cortex Search when it needs context from text, PDFs or other unstructured content. (YouTube)
The agent assembles and formats the response (often with charts or tables) and returns it to the user, all under Snowflake’s governance model. (Snowflake Documentation)
From the user’s perspective this looks like one conversation. Under the hood, it’s an orchestrated, multi-tool workflow that stays entirely inside Snowflake.
6. Why This Matters – Keyrus View
Snowflake’s Agentic AI Framework is not “just another chatbot”. It’s a pattern for building safe, governed AI assistants that:
Run directly on your Snowflake data estate
Combine analytics and knowledge search in one experience
Respect existing security, RBAC and masking policies
Are observable and auditable at every step
For South African organisations dealing with complex regulatory, data-sovereignty and legacy-system constraints, this matters. It allows you to introduce agentic AI:
Without copying data into external vector databases
Without building a custom orchestration layer from scratch
While keeping governance and auditability front and centre
As a Snowflake Premier Partner, we at Keyrus help customers design this architecture in practice: defining the right agents, building semantic models, curating which data is exposed, and aligning the framework with your data governance and analytics strategy. Contact us at sales@keyrus.co.za for hands-on assistance with Snowflake implementation in your organization.