How to Use Agentforce Builder to Build an AI Agent

**Key Takeaways:**

Learn how Agentforce Builder helps you build, configure, test, and deploy an AI agent.
Set up subagents, actions, data, and instructions to give your agent the right capabilities.
Test your agent with real-world requests before activating it and connecting it to customer-facing channels.

AI agents are becoming a part of enterprise workflows, helping businesses and teams automate tasks, retrieve information and take action across their systems.

This is what Salesforce offers, its AI product ‘Agentforce’ allows organizations to build AI agents that work with business data, follow defined instructions, and execute actions through Salesforce capabilities.

However, building an effective agent involves more than creating an agent and adding a few instructions. You need to configure the right topics, connect relevant actions, manage permissions, establish guardrails, and test how the agent responds to real-world scenarios.

That’s too much to be done, right? Well, not really, because everything that is required to build a powerful AI agent are available in this one environment: the Agentforce Builder.

In this guide, we'll walk through how to build an Agentforce agent using Agentforce Builder, from initial setup and configuration to custom actions, testing, security, and deployment. Whether you're a Salesforce professional or someone planning to build AI agents for your workflow, you'll learn how the different components of Agentforce Builder work together to build and prepare Agentforce agents for real-world uses.

What Is Agentforce Builder?

Before we talk about “how to get started with Agentforce,” let’s understand Agentforce Builder first, think of it as a one stop launchpad to build, customize, test, and deploy Agentforce AI agents. It is where you give an agent a role, define what it should do, connect it to the data it needs to perform actions, and configure how it should interact with users.

You can also test the agent against multiple use cases, review how it responds to different requests, and make changes before making it live.

So instead of switching between different tools and settings, you have Agentforce Builder that brings the key parts of agent development together in one place. From creating an agent to preparing it for real-world use, it all happens inside Agentforce Builder.

Prerequisites Before We Start Building Agentforce Agents

Before you start building your Agentforce agent, you need a Salesforce org with the required Agentforce capabilities enabled. For learning and experimentation, a Salesforce Developer Edition org is a practical starting point if the required features are available in that org. Then you have to enable the following settings:

Enable Einstein

Once the org is ready navigate to the Setup, In the quick find box search for Einstein Generative AI and open Einstein Setup. Turn on Einstein from there.

Enable Agentforce Agents

For Winter ’27, the Agentforce platform is enabled by default for orgs that have Agentforce access through their Salesforce edition, SKU, or licensing. New orgs are enabled when they are created, while existing eligible orgs are automatically enabled.

Note: If you’re using a Salesforce Developer Edition org, make sure to use it regularly. If there is no activity for 45 consecutive days, the org may be deactivated.

Creating Your First Agent in Agentforce Builder

Now that your org is ready, it’s time to create your first Agentforce agent.

From Agentforce Studio, open Agentforce Agents. You may already see a default Agentforce agent available in your org.

To create a new one, you can either select New Agent or New from Script. For this guide, we’ll use the New Agent option and build the agent through the Builder interface.

Choose How You Want to Create Your Agent

When you select New Agent, Agentforce gives you two ways to get started. You can either describe what you want the agent to do in natural language or start with a prebuilt template.

Whichever way you choose, you will be asked to give the agent a name and select a user.

If you are exploring Agentforce for the first time, it is worth looking through the available templates. Select Details on a template to understand what it contains and how its components are structured. These templates can give you a useful idea of how Salesforce organizes an agent using subagents and actions.

For this guide, however, we’ll build an agent from scratch. Simply describe what you want your agent to do, provide the required details, and continue.

Welcome to Agentforce Builder

Once you choose a template or provide a description, you’ll land in the Agentforce Builder. This is where you will do most of the work.

Here you can create new subagents, add or configure actions, define instructions, test the agent, and make other required changes to their configuration.

If you started with a template or described your agent in plain language, Agentforce will create some subagents and actions for you. You can anytime edit, delete, or customize these components based on your requirements.

Note: The previous steps were meant to get you to the starting line or the real Agentforce Builder, and this is where you actually define what you want your agent to do.

Before we start creating subagents and actions, let’s first understand the Agentforce Builder interface.

Understand the Agentforce Builder Interface

The Agentforce Builder can be understood through three main areas: the Explorer, the main workspace, and Agentforce Assistance.

1. Explorer

The Explorer is on the left side of the Builder. It gives you an overview of the different components that make up your agent.

Depending on your agent, you can find areas such as:

Settings
Subagents
Actions
Variables
Connections
Data

The Agent Definition section on the left gives you an overall view of the agent and its configuration associated with it.

2. Main Workspace

The center of the Builder is where you work on your agent or get a complete summary of your agent inside out. This part offers Canvas and Script views, allowing you to work with the same agent configuration through a natural-language interface or a more structured, code-based view.

The workspace also includes Preview, where you can interact with your agent as a user would. You can enter requests, see how the agent responds, its reasoning, and inspect the interaction summary to understand which subagents and actions were used in the process.

At the bottom of the Builder’s main workflow is a console for viewing errors and warnings, which can help you identify configuration or syntax issues before activating the agent.

3. Agentforce Assistance

The Agentforce Assistance panel in the right works as a development assistant inside Builder. You can use natural language to ask it to help create, modify, troubleshoot, or refine parts of your agent.

Configure Your Agent in Agentforce Builder

Now that you understand the main parts of Agentforce Builder, it’s time to start configuring your agent.

Create a Subagent (Topics)

Now let's start building the actual capabilities of the agent.

In the Explorer, find Subagents and click on the + icon. There you have three options, you can create a new subagent, add one from the Asset Library, or use an existing agent as a subagent (when building a multi-agent setup).

When creating a new subagent, provide it with a name, description, and reasoning instructions. The description helps determine when the subagent should be used, while the reasoning instructions guide what it should do after it has been selected.

For example, if you are building a customer service agent, you might create a subagent called Order Support that handles questions about orders, cancellations, and delivery status.

Note: Beginning in April 2026, Salesforce renamed agent topics to subagents. There are no changes to the functionality, so you may still see both terms in Salesforce documentation.

Add Actions to Your Subagent

Subagents define the job an agent needs to do, while actions give it the capabilities to perform that job. After you have created a subagent, you can give it these capabilities by adding actions.

Select the subagent you want to work with and add an action, then click add action. Agentforce Builder allows you to create custom actions, create a utility action or add existing actions from the Asset Library. For custom actions, they can be created using Apex, Flow, Prompt Template, and API.

If you’re an individual or a business with complex requirements that need custom actions, talk to our Agentforce AI experts to build the right solution for your use case.

Configure Data for Your Agent

Actions give our agent the ability to perform tasks, while data provides the information it needs to respond and complete those tasks.

Agentforce can work with different types of data, some of the common ones are:

Salesforce CRM data: This is the data already sitting in your org, and your agent can use it through the actions and capabilities you've set up.
Agentforce Data Libraries: If your agent needs to work with unstructured content, like Knowledge articles or uploaded files, a Data Library is the way to go.
External data sources: Data 360 also supports external connectors. You can either pull that data into Data 360 or access it using zero-copy data federation.

For our use case, we're keeping things simple and sticking with the Salesforce data our agent can already access. If yours needs deeper knowledge later, you can always add a Data Library or connect an external source.

Configure Instructions and Guardrails

While creating an agent, it is equally important to ensure security and trust, not just accuracy.

With the topics and actions now configured, it is time to define the security layer that will control how our agent will behave. In Agentforce, these guardrails are configured through topic instructions, which set boundaries, provide context, and guide how the agent makes decisions.

Add Instructions to Your Topics

To secure your agents, open the topic you want to configure in Agentforce Builder and review its instructions.

Use clear instructions to tell the agent what it should do and when it should use specific actions. Salesforce recommends writing instructions as direct rules, such as Always, Never, If...then, or As a first step.

For example, you might configure instructions that tell the agent to:

Always verify the required customer information before performing an action.
Never provide information outside the topic's intended scope.
Ask for clarification when required information is missing.
Escalate the conversation when the request requires human assistance.

You can modify instructions for standard topics or add instructions when creating a custom topic.

Define the Agent's Boundaries

The topics and actions you configure determine what the agent can handle. Keep the agent's scope aligned with its intended business use case and only give it access to the actions and data it needs.

Salesforce also recommends defining clear human handoff criteria and using topic instructions to specify when an interaction should be escalated.

Once the instructions and boundaries are configured, we can test the agent against different scenarios to confirm that it follows the expected behavior.

Preview and Test Your Agent

Once all the configuration is done, select Preview and interact with your agent as a user would.

Try with realistic requests that the agent is expected to handle and observe both the response and the interaction details. Agentforce Builder allows you to inspect which subagents and actions were used during the process, giving you a view into how the agent handled the request.

Don't worry about perfecting the agent in the first attempt. This is an iterative process. Test it, identify where the behavior needs improvement, update the instructions or configuration, then test it again.

Save, Commit, and Activate Your Agent

When you are satisfied with the agent, save your changes and Commit Version. Committing creates a version of the agent that can be activated. After committing the version, select Activate to make the agent available through its configured channels.

At this point, you have gone from an empty Agentforce Builder workspace to a configured and testable agent.

Publish and Deploy Your Agent

Activating the agent makes the configured version available, but the work does not end there. You still need to publish the web deployment, connect the agent to the appropriate routing flow, and add the customer-facing messaging experience.

So, how to deploy an Agentforce agent? The exact steps depend on the type of agent, deployment channel, and your Salesforce setup. For example, an agent can be deployed to a website or used through supported Salesforce and messaging experiences. Here is a quick example of what we will be doing.

Publish the Web Deployment

We will start by publishing the latest version of the web deployment.

From Setup, search for Embedded Service Deployments in the Quick Find box. Open Embedded Service Deployments and select the web deployment. This will be visible by ESA Web Deployment.

Click Publish to publish the latest changes. (This can take some time)

Update the Routing Flow

Next, we are updating the flow that routes requests to your Agentforce agent.

In Setup, search for Flows in the Quick Find box and open the Route to ESA flow. Select the Route to ESA component and set Route to (select your agent)

Then select your Agentforce Agent. Select Save As New Version, save the flow, and then Activate the new version.

If your Agentforce agent does not appear in the routing options, return to Agentforce Builder and confirm that the agent is activated.

Add Embedded Messaging to Your Experience Cloud Site

Now we'll add our agent to our customer-facing Experience Cloud site.

From Setup, we’ll go to All Sites. Find our Experience Cloud site and select Builder.

In Experience Builder, open the Components panel and search for Embedded Messaging. Drag the component onto the page where you want customers to access the agent.

Leave the default settings in place, then select Publish and confirm the publication.

Note: These are the deployment steps for the setup used in this guide. Your deployment process may be different depending on your Agentforce agent, channel, Salesforce configuration, and requirements. If you’re unsure which approach is right for your setup, talk to our Agentforce experts for guidance.

Conclusion

Building your first Agentforce agent is fairly straightforward once you understand what each part of Agentforce Builder does. You create the agent, set up its subagents and actions, connect the data it needs, add instructions, and test it before putting it in front of users.

Agentforce Builder gives you one place to work through all of these steps and make changes as you test the agent.

For simple use cases, a natural language description or an existing template may be enough to get started. But not every project stays simple. Once you need custom actions, data spread across several systems, or multiple agents working together, the setup can get harder to do alone. This is where an experienced Salesforce consultant can help. If you're at that point, we'd be happy to help. At Concretio, our engineers build custom Agentforce agents that fit the way your teams actually work.

Salesforce Updates

https://www.concret.io/blog/salesforce-winter-27-updates

Demystifying Multi-Agent Orchestration in Salesforce Agentforce

Agentforce New Pricing Tiers and Credit Allocations | Salesforce News

Salesforce Koa Explained: NVIDIA Nemotron CRM Reasoning Model (2026)

Frequently asked questions

No. You can describe the agent in plain language, start from a template, and fine-tune it in Canvas view. Script view is there if you want to write Agent Script directly, but you can build a working agent without it.
A subagent defines the job, like handling order questions. Actions are what the agent does to complete that job, such as looking up a record or running a Flow. Subagents set the scope; actions do the work.
Your Salesforce CRM data, Agentforce Data Libraries (for Knowledge articles and uploaded files), and external sources connected through Data 360. Start with the CRM data your agent can already reach and add more only when the use case needs it.
Yes. Salesforce renamed topics to subagents in April 2026. The functionality is unchanged, only the name is new, so you'll still see both terms in older documentation, tutorials, and community posts.
It's a panel on the right side of the Builder that helps you create, edit, and troubleshoot your agent in plain language. Ask it to add a subagent, refine instructions, or fix errors, and it shows the proposed changes for your approval before applying them.

How to Use Agentforce Builder to Deploy Your First Autonomous Agent.

What Is Agentforce Builder?