Snowflake GES-C02: Snowflake for Gen AI Overview Practice Questions

AI Studio and REST APIs are both named entry points for interacting with Cortex, alongside SQL and Snowpark Python. Resource monitors are a cost-governance control for credit consumption, not an interaction/entry-point mechanism. Row access policies are a data-governance control restricting visible rows, also not a way to interact with Cortex itself.

Cortex Agents orchestrate multiple tools — such as Search and Analyst together — for multi-step tasks, which is exactly this scenario. Cortex Search alone can retrieve the policy document but not the structured order history. Cortex Analyst alone can compute from tables but not retrieve document text. A single longer AI_COMPLETE prompt can't itself query separate structured and unstructured sources.

Cortex Agents take a natural-language goal and orchestrate the tool calls needed to satisfy it. A single AI function like AI_CLASSIFY typically takes a column of text or a prompt for a one-shot transformation, not an open-ended goal. EMBED_TEXT_1024 takes text to embed, a narrow input for building a vector, not a goal. VECTOR_COSINE_SIMILARITY takes two vectors to compare, an even narrower mathematical input.

A one-shot classification task is exactly what a single AI function call handles. Reaching for Cortex Agents for a task with no multi-step orchestration adds unnecessary complexity. Cortex Analyst answers questions computed from structured tables, not text classification. Cortex Search retrieves passages; it doesn't classify.

Snowflake Intelligence was renamed to Snowflake CoWork. CoCo is the renamed product formerly called Cortex Code, a different product. Cortex AI Functions is the renamed Cortex AISQL, also unrelated to Snowflake Intelligence's rename. Cortex Guard is a response-safety filtering feature, not a renamed conversational agent.