SnowPro Specialty: Gen AI Question 13
Single answerCortex Agents (Public Preview)A retail company is piloting a customer-support assistant using Snowflake Cortex Agents (Public Preview). The assistant must answer order-status and return-policy questions by combining structured order data in Snowflake tables with unstructured policy documents stored in the company's knowledge base. The team wants to minimize custom orchestration code and let the assistant decide when to use available enterprise data sources during a conversation. Which approach best fits Cortex Agents for this requirement?
- A
Build a Cortex Agent and configure it with access to the relevant structured and unstructured enterprise data sources so the agent can reason over the user request and decide which sources to use
- B
Use AI_COMPLETE with a single prompt that contains only the user's question, because Cortex Agents automatically discover and query Snowflake tables without any additional configuration
- C
Create a semantic model only, because Cortex Agents can use semantic models but cannot work with unstructured documents in a knowledge base
- D
Fine-tune a custom model on order and policy data first, because Cortex Agents require model fine-tuning before they can answer questions grounded in enterprise data
Show answer and explanation
Correct answer: A
Explanation
The best answer is Option 1 because Cortex Agents (Public Preview) are intended for scenarios where an application needs agentic behavior over enterprise data, including selecting among configured data sources to fulfill a user request. This is especially useful when the workflow spans both structured data, such as order tables, and unstructured content, such as policy documents, and the goal is to reduce custom orchestration logic. Option 2 confuses basic LLM completion with agent capabilities. Option 3 incorrectly limits Cortex Agents to structured access only. Option 4 reflects a common but inaccurate assumption that enterprise-grounded assistants require fine-tuning, whereas retrieval and orchestration over trusted enterprise sources are typically the preferred pattern. Candidates should align their design choice with Snowflake documentation describing Cortex Agents as a way to build agents that can reason over requests and use enterprise data sources rather than relying solely on prompt stuffing or model retraining.
- A. Correct.
Correct. Cortex Agents are designed to help with agentic retrieval and answering by reasoning over a user request and dynamically using configured enterprise data sources, including structured and unstructured sources. In this scenario, the requirement is to combine Snowflake table data with knowledge-base content while minimizing custom orchestration. Configuring a Cortex Agent with those sources is the best fit.
- B. Incorrect.
Incorrect. AI_COMPLETE is a text generation function, but it does not by itself provide the agentic orchestration behavior described in the scenario. Cortex Agents do not automatically discover and query enterprise data sources with zero configuration. The team must explicitly configure the relevant tools or data sources the agent can use.
- C. Incorrect.
Incorrect. A semantic model can help with structured data access, but this option is too limited and incorrectly states that Cortex Agents cannot work with unstructured documents. The scenario explicitly requires both structured order data and unstructured return-policy content, which is exactly the kind of mixed-source use case Cortex Agents are intended to support.
- D. Incorrect.
Incorrect. Fine-tuning is not a prerequisite for using Cortex Agents with enterprise data. A common misconception is that grounded enterprise question answering always requires model training on proprietary data. In practice, Cortex Agents focus on using configured data sources and reasoning/orchestration rather than requiring a custom fine-tuned model first.