Overview
Snowflake for Gen AI Overview is 18% of the GES-C02 exam and is the map for the rest of the blueprint. GES-C02 is the current version of SnowPro Specialty: Gen AI and has replaced GES-C01. The domain has two objectives. The first asks you to define Snowflake's Gen AI principles and features: the Cortex models and AI functions that run inside Snowflake, Cortex Search for retrieval over unstructured text, Cortex Analyst for natural-language questions over structured data, Cortex Agents that orchestrate those tools, Snowflake Intelligence as the end-user agent experience, Cortex Code, and bringing your own models through the Model Registry and Snowpark Container Services.
The second objective outlines capabilities in more depth: prompting, vector embeddings and context windows, semantic views and verified queries, cross-region inference, REST APIs, the Model Context Protocol and Cortex Knowledge Extensions. Questions here are mostly about choosing the right capability for a stated need.
How it's tested
Domain 1 questions describe a business need and ask which Snowflake Gen AI capability fits. Expect short scenarios about searching documents, asking questions of tables, combining both through an agent, or hosting a custom model. Read for the data type (structured or unstructured), who the user is, and whether the answer must stay inside Snowflake.
-
Pick the Cortex service
A use case is described and you choose between AI functions, Cortex Search, Cortex Analyst or Cortex Agents.
-
Managed or bring your own
A model requirement is described and you decide between a Cortex-hosted LLM and a custom model deployed through the Model Registry or Snowpark Container Services.
-
Access path
An integration need is given and you select SQL functions, REST APIs, MCP, Snowflake Intelligence or Cortex Code as the interface.
Sales leaders want to type questions such as quarterly revenue by region and get correct numbers from governed tables. That is structured data and the answer must be precise, so Cortex Analyst with a semantic view or semantic model is the fit, ideally with verified queries for common questions. Cortex Search would retrieve text passages rather than compute aggregates, and a raw AI_COMPLETE prompt cannot reliably generate SQL against your schema.
Objectives 2
Official objectives from the Snowflake outline, in outline order.
1.1 Define Snowflake's Gen AI principles and features.The Cortex feature set and the principle of AI next to governed data.
- What it means
- This objective covers the building blocks: Cortex-hosted LLMs and AI functions, optional fine-tuning, Cortex Search for retrieval-augmented generation over unstructured data, Cortex Analyst for text-to-SQL, Cortex Agents, Cortex Code in Snowsight and the CLI, Snowflake Intelligence and its interfaces, and bring-your-own-model options using the Model Registry and Snowpark Container Services.
- What to understand
- Know the one-line purpose of each feature and the data type it targets. Cortex functions are called from SQL or Python and are fully managed, so there is no infrastructure to run. Cortex Search returns ranked passages for RAG. Cortex Analyst turns questions into SQL using a semantic layer. Agents plan across Search and Analyst tools. Custom or open-source models that Cortex does not host go into the Model Registry and can be served on Snowpark Container Services compute pools.
- How it shows up
- A short requirement is given and you name the Cortex feature or the bring-your-own-model route that satisfies it.
- Common confusion
- Treating Cortex Search and Cortex Analyst as interchangeable, when one serves text retrieval and the other structured SQL answers.
1.2 Outline Gen AI capabilities in Snowflake.Deeper capability detail: prompting, embeddings, semantic layers, inference options and integration paths.
- What it means
- This objective goes a level deeper: prompt design, AI functions, vector embeddings and context windows, Cortex Search multi-index queries and access control, Cortex Analyst semantic views, YAML semantic models, verified queries and custom instructions, Cortex Agents, Snowflake Intelligence, cross-region inference, REST APIs, the Model Context Protocol, the Cortex Code CLI and Cortex Knowledge Extensions.
- What to understand
- Embeddings are numeric vectors that let you compare meaning; context windows limit how much text a model can take per call, which is why chunking matters. Semantic views define business metrics and joins so Analyst writes correct SQL, and verified queries anchor common questions. Cross-region inference lets a request run in another region when the model is not local. REST APIs and the Snowflake-managed MCP server expose Cortex to external apps and agents. Knowledge Extensions bring licensed third-party content through the Marketplace as Search services.
- How it shows up
- A scenario hinges on a capability detail, such as needing a model unavailable locally or exposing Search to an external agent.
- Common confusion
- Assuming every model is available in every region without enabling cross-region inference.
High-yield concepts
The terms, rules and formulas this domain turns into questions most often.
-
Cortex AI functions
Managed LLM functions callable in SQL and Python.
- Remember
- No infrastructure; billed per token.
- Tested as
- Core capability questions.
- Trap
- Thinking you must deploy a model first.
-
Cortex Search
Managed hybrid search service for RAG.
- Remember
- Combines vector and keyword search with reranking.
- Tested as
- Unstructured retrieval questions.
- Trap
- Using it to compute aggregates.
-
Cortex Analyst
Text-to-SQL over structured data.
- Remember
- Needs a semantic view or YAML semantic model.
- Tested as
- Structured question scenarios.
- Trap
- Pointing it at raw tables with no semantic layer.
-
Cortex Agents
Orchestrator that plans and calls tools.
- Remember
- Combines Search, Analyst and custom tools.
- Tested as
- Mixed-data scenarios.
- Trap
- Expecting one service to cover both data types.
-
Snowflake Intelligence
End-user agent experience for business users.
- Remember
- Built on Cortex Agents.
- Tested as
- Interface questions.
- Trap
- Confusing it with a developer API.
-
Cortex Code
Snowflake's AI coding assistant.
- Remember
- Available in Snowsight and as a CLI.
- Tested as
- Developer productivity questions.
- Trap
- Mistaking it for a model-hosting service.
-
Cross-region inference
Runs a request in another region when needed.
- Remember
- Disabled by default; set at account level.
- Tested as
- Model availability questions.
- Trap
- Assuming all models exist in every region.
-
Model Context Protocol (MCP)
Open protocol to expose tools to agents.
- Remember
- Snowflake-managed MCP server exposes Cortex tools.
- Tested as
- Integration questions.
- Trap
- Confusing it with the REST API alone.
-
Bring your own model
Custom models via Model Registry and SPCS.
- Remember
- Used when Cortex does not host the model.
- Tested as
- Model choice questions.
- Trap
- Using SPCS when a Cortex model already fits.
-
Cortex Knowledge Extensions
Third-party content shared as Search services.
- Remember
- Delivered through Snowflake Marketplace.
- Tested as
- Data enrichment questions.
- Trap
- Thinking it requires copying the provider's data.
-
Verified queries
Approved question-to-SQL pairs for Analyst.
- Remember
- Improve accuracy on common questions.
- Tested as
- Analyst accuracy questions.
- Trap
- Confusing them with custom instructions.
Side by side
Concepts candidates mix up, compared on the point that decides the answer.
Choosing a Cortex capability
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| AI functions | Row-level LLM tasks in SQL | Process each row or prompt | Using them for search over millions of chunks |
| Cortex Search | Ranked passage retrieval | Unstructured text and RAG | Expecting numeric aggregation |
| Cortex Analyst | Text-to-SQL answers | Structured tables via a semantic layer | Skipping the semantic model |
| Cortex Agents | Plans across tools | Mixed structured and unstructured needs | Building an agent for a single function call |
Managed models vs. your own models
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| Cortex-hosted LLMs | Snowflake-managed models | Serverless, pay per token | Deploying containers unnecessarily |
| Model Registry + SPCS | Your own or open-source models | You manage images and compute pools | Assuming Cortex can host any model |
Ways to reach Cortex
| Concept | Meaning | Key difference | Common trap |
|---|---|---|---|
| SQL and Python functions | Calls inside Snowflake | Simplest for pipelines | Overlooking them for batch work |
| REST APIs | HTTPS calls from external apps | Need authentication tokens | Forgetting authentication setup |
| MCP server | Exposes Cortex tools to agents | Standard protocol for agent clients | Treating it as a model |
Exam traps
Each trap is an answer that looks right. Learn the clue that should change your mind.
-
Search for metrics
- What candidates pick
- Use Cortex Search to answer revenue questions.
- Why it looks right
- Search sounds like the general question tool.
- Why it's wrong
- Search retrieves text passages; exact aggregates over tables come from Cortex Analyst generating SQL against a semantic layer.
Clue to watch for Questions ask for totals, trends or comparisons over tables.
-
Deploy what Cortex already hosts
- What candidates pick
- Package an open model into Snowpark Container Services.
- Why it looks right
- It feels like more control.
- Why it's wrong
- If Cortex already offers a suitable model, the serverless function is simpler and cheaper; SPCS is for models Cortex does not host.
Clue to watch for The model named is in the Cortex catalog.
-
Analyst without semantics
- What candidates pick
- Point Cortex Analyst straight at raw tables.
- Why it looks right
- Text-to-SQL sounds automatic.
- Why it's wrong
- Analyst relies on a semantic view or semantic model that defines tables, metrics and relationships to produce reliable SQL.
Clue to watch for The scenario mentions inconsistent or wrong SQL answers.
-
Region assumption
- What candidates pick
- Call a model that is not offered in the account's region.
- Why it looks right
- Docs list the model as generally available.
- Why it's wrong
- Availability differs by region; cross-region inference must be enabled by an admin before requests can be routed elsewhere.
Clue to watch for An error says the model is unavailable.
-
Agent for everything
- What candidates pick
- Build a Cortex Agent for a single classification task.
- Why it looks right
- Agents are the newest feature.
- Why it's wrong
- A single row-level task like classification is best served by an AI function; agents add value when planning across several tools.
Clue to watch for The task is one well-defined operation per row.
-
Intelligence is an API
- What candidates pick
- Describe Snowflake Intelligence as a developer endpoint.
- Why it looks right
- It is built on agents.
- Why it's wrong
- Snowflake Intelligence is the business-user chat experience; developers integrate through REST APIs, SQL functions or MCP.
Clue to watch for The user is a business analyst, not a developer.
Answer framework
Use these steps to map any Domain 1 scenario to the right Snowflake Gen AI capability.
- Classify the data: structured tables point to Cortex Analyst, unstructured text points to Cortex Search or AI functions, and both together suggest Cortex Agents.
- Decide whether the job is per-row processing, retrieval for RAG, question answering over metrics, or multi-step orchestration across tools.
- Check whether a Cortex-hosted model fits; only move to the Model Registry and Snowpark Container Services when you need a custom or unsupported model.
- Confirm region availability and whether cross-region inference must be enabled, and note any data residency limits the scenario states.
- Pick the access path that matches the user: Snowflake Intelligence for business users, SQL or Python for pipelines, REST APIs or MCP for external apps and agents.
Practice question
Want more like this? The GES-C02 practice test has questions for every domain.
Checklist
0 of 11
Tick each one you can do without notes. Your ticks stay in this browser.
Study strategy
Treat this domain as the vocabulary layer for the exam and study it first, even though it carries only about 18% of the weight. Build a one-page map with each Cortex capability, the data it works on, the user it serves and how it is invoked. Practice by reading short use cases and naming the capability in a few seconds: retrieval over documents, metric questions over tables, multi-tool assistants, or custom model hosting.
Next, go one level deeper on the details that objective 1.2 lists. Learn how embeddings and context windows shape chunking, what a semantic view contains, how verified queries and custom instructions steer Cortex Analyst, and what cross-region inference changes. Read the Snowflake documentation pages for Snowflake Intelligence, the managed MCP server, Cortex Code and Cortex Knowledge Extensions, because they are new in GES-C02 and absent from many older courses.
In a trial account, run one AI_COMPLETE call, create a small Cortex Search service and try Cortex Analyst with a simple semantic view. Then move on to Domain 2, where these capabilities are applied in detail.
FAQ
How much of the SnowPro Gen AI exam is Snowflake for Gen AI Overview?
Snowflake for Gen AI Overview is 18% of GES-C02, roughly 10 of the 55 questions. It has two objectives: defining Snowflake's Gen AI principles and features, and outlining Gen AI capabilities in Snowflake.
Is GES-C02 the current exam version?
Yes. GES-C02 replaced GES-C01 in 2026. It reorganizes the content into four domains and adds topics such as Snowflake Intelligence, Cortex Code, MCP and Cortex Knowledge Extensions while removing the older Document AI product.
What is the difference between Cortex Search and Cortex Analyst?
Cortex Search retrieves relevant text passages from unstructured data for RAG. Cortex Analyst converts natural-language questions into SQL over structured tables using a semantic view or semantic model.
When would I use Snowpark Container Services for Gen AI?
When you need a model Cortex does not host, such as a fine-tuned open-source model or custom code, you log it in the Model Registry and serve it on SPCS compute pools, often with GPUs.
What is cross-region inference?
An account-level setting that lets Cortex process a request in another Snowflake region when the chosen model is not available locally. It is off by default and an administrator must enable it.
What is Snowflake Intelligence?
The business-user experience for chatting with data, built on Cortex Agents. Users ask questions in natural language and the agent draws on Cortex Analyst, Cortex Search and other configured tools.
How does Domain 1 relate to the other domains?
It introduces the capabilities that Domain 2 applies, Domain 3 secures and budgets, and Domain 4 uses for documents, so its vocabulary appears throughout the exam.
Other Artificial Intelligence & Machine Learning certifications
Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.
Test your Domain 1 knowledge
You have mapped Cortex AI functions, Search, Analyst, Agents, Snowflake Intelligence and bring-your-own models. See whether you can pick the right capability under timed conditions.
Keep going
- Training Training Built on your Weak areas
- Practice test Questions by domain, with explanations
- Mock exam Full length, timed like the real exam
- Readiness test Find your weakest domain in minutes
- Cheat sheet Key terms and formulas on one page
- Ask the GES-C02 AI tutor Get any concept explained your way