Salesforce Certified Agentforce Specialist 1-10

In Testing Center, it is important to run tests safely under conditions close to production data.
For that reason, Salesforce recommends using a sandbox environment that is cloned from production.
A sandbox copies production data while remaining isolated from the production environment, so you can perform realistic validation without affecting live operations.
Option A’s Developer org has insufficient data and does not enable realistic testing, and Option B’s direct testing in production is inappropriate because of the risk of data corruption and security issues.
Therefore, C is correct.
Agentforce Testing Center

Using Einstein Studio’s BYO-LLM (Bring Your Own Large Language Model) capability, you can natively integrate an external large language model within your Salesforce environment.
This allows it to work together with Prompt Builder, enabling flexible generative AI implementations that leverage an external LLM.
Option A’s connection via Apex requires heavy implementation effort and is not recommended, and Option C’s Flow or external service is mainly for data integration purposes and is not optimal for leveraging an LLM.
Therefore, using the standard BYO-LLM capability is the most efficient and recommended configuration.
Bring Your Own Large Language Model

Prompt Builder provides a mechanism for dynamically retrieving data through Data Cloud related lists and prompt-triggered flows.
This makes it possible to leverage context-aware generative AI based on unified customer data.
Direct DMO access from a Flex template, as in Option A, is not supported, and using only an external API, as in Option C, is unnecessary.
Therefore, combining the standard related lists and flows is the optimal and recommended approach.
Grounding with Flow Merge Fields

For the sales team to use the prompt template, granting the permission to use templates (Prompt Template User) is a prerequisite.
In addition, enabling the Sales Email feature makes it possible to link email workflows with the prompt template.
The Manager permission in Option B is for creating and managing and is not needed for usage, and the Data Cloud Admin permission in Option C is unrelated to this feature.
Therefore, the combination of granting the appropriate permission and enabling the feature is the best practice.
Enable Prompt Builder

To use a field generation prompt template, the target field must be explicitly associated with the prompt template on the Lightning page layout.
If this setting is not done, the sparkle icon that indicates AI generation does not appear, and the field is treated as a normal field.
Also, activating the template alone is not enough; adding the field to the page layout and configuring the association are required.
Option A’s re-activation is unnecessary, and Option B is incorrect because the Case object is supported.
Therefore, C is correct.
Add a Field Generation Prompt Template to a Lightning Record Page

Because this requirement calls for answering based on the PDF internal policy documents, it is important to first build a knowledge foundation (Data Library) that the AI can reference.
Using the Agentforce Data Library, you can upload PDFs, index them, and provide them to the AI in a searchable form.
This achieves Retrieval-Augmented Generation (RAG), so the AI can answer based on the company’s approved data rather than the LLM’s general knowledge.
Option B is too broad in scope, and Option C is insufficient as a concrete configuration procedure.
What Is a Data Library?

In Model Playground, the main hyperparameters that can be adjusted to control the behavior of the generative AI model are temperature, frequency penalty, and presence penalty.
Temperature controls the diversity of the output, frequency penalty suppresses repetition, and presence penalty encourages the generation of new topics.
These allow you to fine-tune the nature of responses.
On the other hand, Top-k sampling is not provided in Salesforce’s standard UI, and the number of output tokens is not a primary adjustment parameter.
Therefore, the correct combination that Agentforce can change is these three hyperparameters.

Create a Model Configuration

This requirement calls for targeting only published knowledge articles while also showing the user the basis for the response.
Therefore, enabling “Use published knowledge articles” limits the indexing target to only published content.
In addition, enabling “Show sources” makes it clear which knowledge articles the LLM’s response is based on.
The data categories in Option A are for classification and are insufficient for controlling indexing, and Option B does not meet the requirement because it cannot display sources.
Therefore, C is correct.
Use Salesforce Knowledge

Action instructions function as a guide for the reasoning engine to determine which action to execute and when.
Specifically, by defining the action’s purpose, input conditions, and execution timing, they correctly connect the user’s intent with the appropriate action.
This allows the AI agent to select the optimal processing according to the situation.
On the other hand, the LLM temperature setting is managed at the system level and is not controlled by the action instructions.
It is also important that this is a setting for internal logic rather than an explanation for users.
Best Practices for Agent Action Instructions

A prompt template links to the LLM it references by name or ID. Therefore, if the same LLM configuration (name and ID) does not exist in both the sandbox and production environments, the dependency cannot be resolved and a deployment error occurs.
In particular, when using BYO-LLM, the same configuration must be created in advance across environments. A mismatch in the LLM name or configuration is the most typical cause, so it is important to ensure consistency before migration.Option A is not the fundamental cause, and Option B is incorrect because BYO-LLM is available in production.
Prompt Builder