Understanding the Snowflake Gen AI Ecosystem

Those who followed last blog for exam breakdown, hope you have landed here as well. If you have not checked out exam breakdown part, I suggest go through it here.

Now it’s time to actually start learning. Domain 1 tests you should know overall features of Snowflake Gen AI. Please take a closer look at the below diagram as per the study guide they want you to know all these topics.

Domain 1 — Topics

I will divide this topic in 2 parts.

Part 2A — Understanding the Snowflake Gen AI ecosystem

Part 2B — Understanding Snowflake’s Gen AI capabilities

This domain is not about a hands-on practice but it is about you understand all the features of Gen AI overall. This domain will test you on how you are aware of all the capabilities and how you take the design decisions.

Now, let’s understand the Gen AI ecosystem inside Snowflake and where each capability fits.

The important capabilities I want to focus on are:

1. Snowflake Cortex
2. Cortex AI Functions
3. Cortex Search
4. Cortex Analyst
5. Cortex Agents
6. Cortex Code
7. Snowflake Intelligence → now Snowflake CoWork
8. Bring your Own Model

Let’s take them one by one.

1. Snowflake Cortex

The first thing to understand is Snowflake Cortex. Snowflake Cortex is not just one function or one model. It is a collection of AI capabilities available directly within Snowflake.

These include capabilities such as:

• Cortex AI Functions
• Cortex Analyst
• Cortex Search
• Cortex Agents
• Cortex Fine-tuning
• Cortex Code

Snowflake’s current AI/ML documentation groups these capabilities under the Snowflake AI & ML platform.

2. Cortex AI Functions — AI directly in SQL

This is one of the most important concepts to understand.

Cortex AI Functions allow you to apply AI to your data using SQL.

For example, suppose you have a column ‘Review’ in ‘CUSTOMER_REVIEWS’ table:
REVIEW
— — — — — — — — — — — — — — —
“Product is excellent”
“Delivery was very slow”
“Great product but poor packaging”

You can use an AI function to determine sentiment.

SELECT
REVIEW,
AI_SENTIMENT(REVIEW)
FROM CUSTOMER_REVIEWS;

You don’t need to build a Python application. You can apply AI as part of your Snowflake data pipeline. Snowflake’s Cortex AI Functions support use cases including classification, extraction, sentiment analysis, translation, embeddings, summarization and more.

Some examples worth knowing:

We will cover all of these in details in 4th Blog.

Meanwhile check this out.

3. Cortex Search

Now let’s say your company has thousands of documents:

• HR Policies
• Product Manuals
• Customer Support Tickets
• Contracts
• FAQs
• Knowledge Base

And someone asks:

“What is our company’s maternity leave policy?”

This is not really a SQL analytics problem.

You need to search through unstructured text.

That’s where Cortex Search comes in.

Cortex Search provides low-latency search over Snowflake data using a combination of keyword and vector/semantic search. It is commonly used for enterprise search and RAG applications.

This is the basic RAG pattern.

Check out more here.

4. Cortex Analyst

Now let’s change the scenario.

You have:

SALES
— — — — — — — — — — — — — — — —
ORDER_DATE
CUSTOMER
REGION
PRODUCT
REVENUE

A business user asks:

“What was our revenue in India last quarter?”

The user doesn’t know SQL.

This is where Cortex Analyst comes in.

Cortex Analyst converts natural-language business questions into SQL and answers questions over structured Snowflake data.

Cortex Analyst needs to understand the business meaning of your data.

For example:

REVENUE = SUM(SALES_AMOUNT)

CUSTOMER = Customer Name

A semantic view provides this business-friendly layer.

Snowflake describes semantic views as a layer that defines logical tables, dimensions, facts, metrics and relationships so that Cortex Analyst can understand how the business thinks about the data.

Check out here.

5. Cortex Agent

Now imagine a much more complex question:

“Why did our sales decline in last quarter? Check the sales numbers and also look at customer complaints.”

Now we need:

Structured data AND Unstructured data

This is where Cortex Agents become useful.

A Cortex Agent can reason over a request, plan the work and use tools such as:

• Cortex Analyst
• Cortex Search
• Code execution
• Custom tools
• MCP connectors
• Other agents/toolsets

Snowflake describes Cortex Agents as a managed agentic platform that can bring structured and unstructured data together in a governed workflow.

Get Minal Bhatkar Satpute’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

In this situation, Agent might decide:

Step 1 → Check sales numbers
↓
Cortex Analyst
Step 2 → Search customer complaints
↓
Cortex Search
Step 3 → Compare findings
↓
Agent reasoning
Step 4 → Generate answer

This is why an Agent is different from simply calling an LLM.

An agent can orchestrate tools to accomplish a task.

Check out here.

6. Cortex Code

Cortex Code is an AI-driven agentic assistant for tasks such as:

• SQL development
• Python development
• Data exploration
• Data engineering
• ML workflows
• Snowflake administration
• Agent development

Cortex Code is focused on helping developers work with Snowflake using natural language and AI-assisted capabilities.

The study guide specifically includes:

• Cortex Code in Snowsight
• Cortex Code Command Line (CLI)

You can try it out here.

The basic difference is:

Cortex Code = AI assistant for building, exploring and working with Snowflake

Whereas:

Cortex Agent = AI application/agent that uses tools to answer or execute business tasks

7. Snowflake Intelligence — now Snowflake CoWork

Snowflake Intelligence was introduced as an enterprise intelligence agent where business users could ask questions in natural language and get insights from their enterprise data.

As of the current product naming, Snowflake Intelligence is now Snowflake CoWork. Snowflake describes CoWork as a personal work agent that can reason across enterprise data and take action through connected tools.

Think of it from a business user’s perspective:

The user doesn’t necessarily need to know SQL, Python, Cortex Search, Semantic Views, APIs

The goal is to provide a conversational interface over governed enterprise context.

Different ways to interact with Snowflake Gen AI

• AI Studio
• SQL
• REST APIs

This is important for certification preparation because the study guide doesn’t only focus on the AI capabilities themselves — it also expects us to understand the different ways we can interact with them.

Check out here for more.

8. Bringing your own models into Snowflake

Then introduce BYOM without getting too deep.

So far, we’ve mostly talked about capabilities provided by Snowflake.

But what if the model we want to use isn’t one of those?

This is where bringing your own models into Snowflake becomes important.

The study guide points us toward two important areas:

Snowflake Model Registry

Snowpark Container Services

These are worth studying because they represent a different approach from simply using Snowflake’s managed AI capabilities.

I’ll cover the implementation details later, but for now the important distinction is understanding managed Snowflake AI capabilities vs. bringing your own model into the platform.

How do all these capabilities fit together?

This is probably the most important diagram to remember from Domain 1.

Remember the problem each solves:

What I would study for the exam

For Domain 1, don’t just read the product overview.

Make sure you can answer these questions:

Cortex

• What is Snowflake Cortex?
• Which capabilities are part of Cortex?

AI Functions

• Which function solves which problem?
• Which functions work with text/images/documents?
• What is the difference between generation, classification, extraction, embedding and similarity?

Cortex Search

• Structured vs unstructured?
• What is hybrid search?
• How is Cortex Search used for RAG?
• How does it work with Cortex Agents?

Cortex Analyst

• What is Text-to-SQL?
• Why do we need a semantic view?
• What are dimensions, facts and metrics?
• What are verified queries?
• How does Semantic View Autopilot help?

Cortex Agents

• What is an agent?
• What tools can an agent use?
• How does an agent use Analyst + Search?
• When should you use an Agent instead of a single Cortex capability?

Cortex Code

• What is Cortex Code?
• Snowsight vs CLI?
• What types of development tasks can it help with?
• How does it work with Snowflake context?

Snowflake Intelligence / CoWork

• What is it?
• Who is the target user?
• How does it provide a conversational interface over enterprise data?

Stay tuned for the next blog.