Get Up to Date on Agentforce Concepts

Learning Objectives

After completing this unit, you’ll be able to:

Manage Agentforce user security.
Test an agent using the Testing Center.
Deploy an agent from sandbox to production.
Articulate the steps for creating a test.

Maintain Your Certification

If you hold the Agentforce Specialist certification, keep in mind that you need to complete this module by the due date to maintain your certification.

Interested in learning more about getting certified? Check out the Agentforce Specialist certification.

While anyone can earn this badge, this module is designed for those who hold the Agentforce Specialist certification.

Protect the Integrity of Your Certification

The quality of our certification exams and the value that they provide are our highest priority. Protecting the security and confidentiality of our exams is essential to providing our customers with certifications that are respected and industry-leading.

As a participant of the Salesforce Certification Program, you’re required to accept the terms of the Salesforce Certification Program Agreement. Please review the Salesforce certification exam-taking policies in the Salesforce Certification Program Agreement and Code of Conduct Trailhead Help article for more details.

Salesforce introduced great feature enhancements over the past year. Take a look at some of the more important ones.

Control Agent Access to Topics and Actions with Filters

As of March 2025, you can now create filters for topics and actions in Agentforce, ensuring agents can access them only when specific conditions are met. Previously, agents had access to all added topics and actions without restriction. Now, agents’ access to topics and actions can be controlled using filter and context variables.

Filters help you securely manage agent access to sensitive or conditional workflows.

This is especially useful for Service agents on customer channels. Once you’ve configured and customized the Customer Verification topic to align with your security requirements, you can create filters to control access to topics and actions that perform tasks on a user’s behalf. For example, you might unlock a Case Management topic only when the user’s case status is Escalated, or allow an agent to initiate a return only if the user’s order was received within the past 30 days.

To get started with using filters:

Add the variables to your agent via the Context panel. These serve as conditions in your filters.
Go to the Filters tab and create a filter.

Once you’ve created your filter, here’s how you can apply it to a topic or action.

In the Topics panel, open the Details page for the topic or action you want to control. Apply the filter. Once you do, it’s active for that topic or action within this specific agent only.
Once the filter conditions are met, the agent can access and use the filtered topic or action.

This update applies to Lightning Experience in Enterprise, Performance, Unlimited, and Developer editions.

Secure Agents with Customer Verification

In Agentforce, security isn’t just about authentication. It’s about controlling access to actions within the agent experience. That’s where Customer Verification topics play a key role, ensuring that only verified users can initiate sensitive conversations or perform high-risk actions through the agent.

You can define which topics and actions require verification, tailoring access based on your organization’s specific security policies. For example, if your company offers booking services through a self-service channel, you might want to verify a user’s identity before allowing them to do things like change a reservation, access payment details, or request a refund.

Here’s how it works. If an unverified user attempts to access a protected topic or action, the agent automatically launches the Customer Verification topic. From there, the agent:

Requests the user’s email or username.
Sends a one-time passcode (OTP) to the user via email.
Verifies the user when they successfully enter the OTP.

By adding this step, you greatly reduce the risk of unauthorized access or fraudulent changes.

Test an Agent Using Testing Center

Before launching your generative AI-powered agent into the world, you need confidence that it works correctly in every situation. That’s where Agentforce Testing Center comes in, a new feature that allows you to evaluate agent accuracy and reliability at scale by:

Running batch tests using a simple CSV template
Using generative AI to create test scenarios
Testing up to 1,000 cases per test
Running up to 10 test jobs in a 10-hour window

This results in faster development cycles and more reliable agents that are ready to meet customer needs from day one.

Always run tests in a sandbox environment. Testing agents in production can modify CRM records, and testing consumes API requests and credits.

The Testing Center is designed for ease and flexibility. You can either upload a CSV file using a prebuilt template with test data, or generate test scenarios with gen AI. This is especially helpful for common Salesforce objects or knowledge-based responses. The results highlight successes and pinpoint failures, helping you fine-tune the agent before go-live.

Create and Run a Test

When creating your own test scenarios, start with the provided CSV testing template. Make sure to include values in the Utterances column and at least one additional column. For expected values, use the API names—not the labels—of topics and actions. You can find these API names in Agentforce Builder. Keep in mind that any empty fields in your test file are treated as failures.

From Setup, in the Quick Find box, search for and select Testing Center. Or from Agent Builder, click Batch Test.
Click New Test.
Give your test a name.
Select the agent and add a description.
Click Next.
(Optional) Choose to include context variables (like user profile, location, or channel) to simulate different user environments. You can also choose to include conversation history.
Click Next.
You can generate test cases using AI or use your own scenarios. For AI-generated test cases, you can create test cases based on relevant topics and actions or based on the data available to that agent in Agentforce Data Library. Select one of the following:

Topics and Actions: Generate up to 1,000 test cases by selecting topics to test. Add a description to guide the AI on purpose and scope, then click Next.
Agentforce Data Library: With a data library assigned, generate up to 100 test cases based on the agent’s knowledge. Add a description and select topics, then click Next.
Upload CSV: Use the testing template to upload up to 1,000 custom test cases. Include data in the Utterances column and at least one other. Use API names (not labels) for topics and actions—find these in Agentforce Builder.

Click Generate Test Cases to finalize and create your test set.
To re-run a test, locate it in the test list and select Run Again from the dropdown menu.

Gen AI can generate test cases for standard objects like Account, Lead, Opportunity, and Contact.

With clear instructions in the description, it can also create test cases for custom objects and the Answer Questions with Knowledge action.

For custom objects, include a description such as: “Generate test cases for [custom object name]. Sample utterances include: [sample utterance related to the custom object].”

To generate test cases for the Answer Questions with Knowledge action, use a description like: “Generate test cases for the Answer Questions with Knowledge action. Use the following knowledge articles to generate the tests: [title of an article], [title of an article].”

Providing specific examples in the description helps gen AI create more accurate and relevant test cases.

Agentforce Testing Center gives you a powerful, scalable way to test your agent before it ever talks to a customer. It’s currently available in Lightning Experience and Enterprise, Performance, Unlimited, and Developer Editions. Required add-on licenses vary by agent type. Whether you’re testing dozens or thousands of scenarios, it helps you identify what’s working—and what isn’t—so you can refine your agent and deliver an experience your users can trust.

Deploy Agents with Change Sets

Once your agent is ready for production, deploy it from your sandbox to a production org. This is done using Salesforce change sets—a standard deployment tool for configuration metadata. Change sets are available in both Salesforce Classic and Lightning Experience. They are also available in Enterprise, Performance, Professional, Unlimited, and Database.com Editions.

To work with change sets in Salesforce, specific user permissions are required.

To edit deployment connections, users need both the Deploy Change Sets and Modify Metadata Through Metadata API Functions permissions.
Using outbound change sets requires the Create and Upload Change Sets permission.
Deploying inbound change sets requires both Deploy Change Sets and Modify Metadata Through Metadata API Functions.

When you want to move customizations from your current org to another, you create an outbound change set. Once it’s sent, the receiving org views it as an inbound change set. Sending a change set between two orgs requires a deployment connection, which can only exist between orgs affiliated with the same production org—for example, between a production org and its sandbox, or between two sandboxes that originated from the same production org.

You can also use change sets to deploy a prompt template, which is a directive to help guide agents through various scenarios. This is available in Lightning Experience, Enterprise, Performance, and Unlimited Editions with the Einstein for Sales, Einstein for Platform, Einstein for Service Add-On, or Agentforce foundations.

Create and Upload a Change Set from the Source Sandbox

Log in to your sandbox environment.
From Setup, in the Quick Find box, search for and select Outbound Change Sets.
Click New to create a new change set.
Provide a meaningful name and description, then click Save.
In the Change Set Components section, click Add to start selecting components.
Choose the appropriate Component Type (for example, Generative AI Function Definition, Generative AI Planner Definition, Generative AI Plug-in Definition, Generative AI Prompt Template), select the specific items you want to deploy, and click Add to Change Set.
In the Change Set Detail section, click Upload, choose the destination org (usually Production), and click Upload again.
You’ll receive a confirmation email once the change set has been successfully uploaded.

Deploy the Change Set in the Destination Production Org

Log in to the target production org.
From Setup, in the Quick Find box, search for and select Inbound Change Sets.
Under Change Sets Awaiting Deployment, click the name of the change set you uploaded.
Click Validate to check for any issues. Make sure your org meets the minimum 75% code coverage requirement for successful deployment.
Once validation passes, click Deploy to complete the deployment process.

Tip: You can repeat the validation step before deployment to ensure everything is still aligned and ready for deployment.

Agent Versioning: Work Safely Without Interruptions

Launching a generative AI-powered agent into production isn’t just about making it live—it’s about ensuring reliability, preserving flexibility, and minimizing risk. Whether you’re releasing a new capability or rolling back to a more stable version, Agentforce gives you tools to manage agents with control and confidence—including agent versioning.

Agent versioning lets you test updates without disrupting your active version. And when you’re ready to go live, Salesforce change sets make it easy to deploy your agent across environments. You can create up to 20 versions of an agent, and only one version can be active at a time. This gives you the flexibility to iterate and experiment while keeping production stable.

The number of active agent versions you can have at the same time depends on the agent type. For instance, only one active version is allowed for the Agentforce (Default) type.

To manage agent versions:

Go to the Agentforce Agents Setup page, and click the dropdown next to the agent you want to update.
Choose Clone Version to create a new version, or select Delete Agent Version to remove an existing one.

You can easily switch between different agent versions in both Agentforce Builder and on the agent’s Details page.

Think of agent versioning like software version control. You can create a new version to:

Test changes safely without impacting the current production version.
Stage enhancements before deploying them.
Save backups in case you need to revert.

Permissions Required

To manage agent versions, you need the following permissions.

User Permissions Needed

To manage versions of Agentforce (Default):

Manage AI Agents AND Manage Agentforce Default Agent

Customize Application

To manage versions of other AI agent types:

Manage AI Agents AND the required permissions for your agent type