Explore Agentforce

Learning Objectives

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

Define each component of Agentforce.
Describe what a standard action is.
Explain how a custom action is built.
Summarize how actions fit into topics.

A Look Under the Hood

Before Linda starts setting up Agentforce, she wants to get up to speed on the basics of the feature. So let’s join her as she explores the different components and learns what makes Agentforce tick.

How Agentforce Works

Agentforce has four basic components: the assistant, topics, actions, and the reasoning engine.

Assistant

In the previous unit, you learned that Agentforce is a trusted conversational AI agent seamlessly built into the Salesforce interface. Currently, you can customize and launch one assistant for your employees, and that assistant is available in the flow of work in Salesforce.

Your assistant has the capability to perform business tasks on behalf of the users in your Salesforce org. But how exactly does that magic happen? That’s where actions come in.

Actions and Topics

Actions are how an assistant gets things done. An assistant includes a library of actions, which is basically a set of tasks the assistant can do, such as summarizing information, getting answers from a knowledge base, or drafting emails. Every action in an assistant is assigned to a topic.

Topics are a layer of organization and customization that help your assistant make more accurate decisions and generate more relevant, predictable responses. If you peek inside a topic, you’ll find a collection of related actions and instructions telling Agentforce how to use them.

When a user enters a question or request, instead of searching through a flat list of all actions assigned to it, an assistant selects a relevant topic and then launches one or more actions included within that topic. This keeps your assistant focused on the actions and data that are most relevant to the current conversation. Another way to think of it: a topic is the job to be done (Manage an Account) and actions are the tasks related to that job (Generate an Email, Log Calls, Create a To-Do List). Together, they help the user accomplish their goal.

Now that you understand how topics and actions work together, we can dive a little deeper into the actions themselves.

Types of Actions

Salesforce provides some standard actions right out of the box. So after enabling the feature, your assistant is immediately ready to help users with many common Salesforce tasks. But you can also create custom actions to give your assistant additional abilities, so that it can assist with tasks specific to your business. Let’s take a closer look at these two types of actions.

Some Agentforce actions are in beta and have limited functionality, as further described in the documentation. Including them in an assistant is part of the Services and will consume Einstein Requests if enabled and used.

Standard Actions

Standard actions are provided by default with Agentforce. Some standard actions are available to all users who have permission to access Agentforce. Other standard actions were built to work with specific clouds or products, so they require an additional license.

Below are some of the standard actions that are included with Agentforce, a brief description of what they do, and an example of a user request that might trigger the action. Some of the standard actions are critical for the basic functionality of the assistant, and those are considered system actions. System actions can’t be removed from your assistant.

Action Name	What It Does
Identify Record by Name (system action)	Searches for Salesforce records by name and returns a list of matching records IDs. For example: “Show me the Acme records.”
Identify Object by Name (system action)	Interprets the user’s input to decide which object the user is referring to, then returns the object’s name so additional actions can be taken. For example, if the user enters, “List the opportunities for the Acme account” in the assistant chat window, the action identifies that user is requesting information related to the Account object and Opportunity object.
Query Records (Beta)	Finds and retrieves Salesforce records based on the user’s request and specific conditions, such as the values of fields. For example: “Find all open opportunities set to close this quarter sorted by created date.”
Query Records with Aggregate (Beta)	Answers aggregation questions about Salesforce data, such as count, sum, max, min, or average. For example: “How many opportunities were created in the past 5 days?”
Summarize Record	Summarizes a single Salesforce CRM record. For example: “Create a summary for the Acme deal.”
Draft or Revise Sales Email	Creates a sales email draft or revises the latest generated email based on the user’s input. For example: “Help me write an intro email to Steve from Acme.”
Answer Questions with Knowledge	Answers a question from a user based on information from relevant knowledge articles. For example: “What is the policy for returns over 30 days?” (Requires a Knowledge license.)

As you can see, standard actions are a great start. They give your assistant a set of useful tasks it can accomplish for your employees, and even more standard actions will be available in future Salesforce releases. To see all the standard actions out there, check out the documentation.

If you want to customize your assistant so that it can help users with processes and workflows specific to your business, you can create custom actions.

Custom Actions

The good news about custom actions is that you don’t have to create them out of thin air. In fact, custom actions are based on Salesforce technologies you already know and love.

When you create a custom action, you build it on top of existing platform functionality that you want to make available in Agentforce, such as invocable Apex classes, autolaunched flows, and prompt templates. It’s an awesome way to get more mileage out of your current Salesforce Platform capabilities.

The dropdown list of available action types for a new custom action.

For example, you can use flows to connect to MuleSoft APIs or use Apex or flows to connect to third-party APIs. You can also use Apex or flows to access engagement data, website data, or third-party data through Data Cloud. With custom actions, you can make that functionality available in Agentforce, which unlocks a ton of value and use cases.

Use Cases for Custom Actions

Wondering how your organization might use custom actions? The possibilities are endless, so it depends on the unique needs of your business. Linda has been brainstorming ways Cloud Kicks might take advantage of custom actions. Here are some of the use cases she came up with.

Get order details.
Initiate an order return.
Make product recommendations.
Check inventory levels.
Create and modify invoices.
Book sales meetings.
Gauge customer sentiment.
Create marketing materials.
Log internal IT tickets.

Even though Linda is brimming with ideas, she decides to start small and roll out Agentforce to a small group of users on the sales team to get their feedback first. That way she can test the feature with the standard actions before getting fancy with use cases for custom actions.

To learn more about custom actions, see the links in the Resources section.

Reasoning Engine

Topics and actions are pretty powerful, right? They’re the building blocks of an assistant, the animating force behind your new AI agent. But how does an assistant know when to launch these topics during conversations with an end user? Let’s meet the reasoning engine behind Agentforce.

You can think of the reasoning engine as the conductor of an orchestra. The conductor keeps time and guides a group of musicians to coordinate their individual performances. Similarly, Agentforce’s reasoning engine orchestrates how topics and actions carry out a user’s request.

When a user launches Agentforce and starts a conversation, they want to ask a question or enter an instruction. Behind the scenes, the reasoning engine works with the LLM to carry out the request. Here’s what it does.

Interprets the user's request and classifies the utterance into a topic.
Iteratively builds plans for accomplishing the user's goal.
Finds and launches the right topics and actions to achieve the goal.

Time for Action

Linda’s feeling more confident now that she knows how the feature works, and she’s ready to take Agentforce for a spin. In the next unit, she learns how to enable and customize an assistant in a Salesforce org.

¿Necesita ayuda?

Descubrir más