How Salesforce Agentforce Operates: Unlocking the Magic

Agentforce was unveiled as a groundbreaking suite of autonomous AI agents that help employees handle service, sales, marketing, and commerce tasks, driving exceptional efficiency and customer satisfaction. It enables businesses to scale their workforces on demand with a few clicks. Agentforce’s unlimited AI agents can analyze data, make decisions, and act on tasks like answering customer service inquiries, qualifying sales leads, and optimizing marketing campaigns. By combining the power of Salesforce’s CRM with the specific needs of the organization, Agentforce enables agents to have a 360-degree view of their customers while automating manual tasks and improving overall operational efficiency.

How Does Salesforce’s Agentforce Work?

Salesforce says that agents can be built with clicks, without codes. But what goes behind the actual work? What are the building blocks that we will use to configure our agents? How does the process of building an agent work? We are sure you all had these questions in mind. Agentforce runs on four fundamental Components – Agentforce Agents, the Atlas Reasoning Engine, Agent Builder, and the Partner Network.

Agentforce Agents: These are pre-built solutions that can take input from a user and translate that input into a series of actions based on instructions that you can configure and enhance.

Agent Builder: It is the new low-code tool for Agentforce, which includes Prompt Builder and Model Builder. It lets users tap into existing tools like Flows, prompt templates, Apex, and APIs to configure their Agent.

Atlas Reasoning Engine: It serves as the brain of Agentforce agents. It allows users to automate agents’ actions to respond naturally and adapt to situations so they’re faster, more multi-turn, and more capable.

Partner Network: The partner companies extend Agentforce with new agent actions, and pre-built agent templates, ensuring consistent and responsible AI agents that can plan and act across multiple systems on behalf of users.

Let’s navigate deeper through the main elements of what makes an agent function.

Agent Description

Build your agent using the Agent Builder, a feature within Agentforce Studio. The configuration process is crucial and serves as the starting point. It’s designed to be intuitive, allowing you to use natural language, similar to how you’d converse with someone. As Salesforce puts it, “If you can describe (dream) it, Agentforce can make it happen.” In the Configure Agent menu, you can define the avatar, agent name, and description with channels that represent the methods through which humans and agents interact, such as email, voice, WhatsApp, and more.

There are four sections to this interface:

The left-hand sidebar to navigate between agent settings
The left panel shows the settings page you’re currently on.
The right-hand panel enables you to start a conversation to test-drive your agent.
The middle panel, where the test-drive output will be displayed

Agent Topics

Topics serve as the foundational elements for agents, defining the range of tasks they can perform. You can think of topics as categories of information—for instance, an “order management” topic might allow the agent to access order history and modify product specifications. In the Dreamforce ’24 keynote demo, the agent successfully provided responses and solutions for some customer inquiries, but certain issues remained unresolved because the agent wasn’t equipped with the necessary topics. This limitation reflects the concept of guardrails, which outline what the agent is not permitted to do.

Agent Actions

In addition to topics, actions are another key component of agents. These actions are linked to topics and function the same in Agentforce as they do in Einstein Copilot. Essentially, actions represent the tasks an agent can perform to execute a plan. When you activate Agentforce, a wide range of standard actions are available immediately. If you’re aiming for a quick solution, standard actions are a great starting point. However, if you want to customize actions using your data or objects—or if you need to address a unique use case that standard actions can’t cover—you can create custom actions. These custom actions can be built using Apex, Flows, Prompts, or Service Catalog items (currently in beta).

Atlas Reasoning Engine (Testing Your Agent)

In the Agentforce Builder, the center and right-hand panels allow you to test your agent in real time. This brings the Atlas Reasoning Engine into view within the user interface. When you provide a prompt, you can follow the agent’s reasoning process step-by-step:

Identifies the relevant topic.
Executes actions (such as querying the CRM database via flow).
Retrieves the correct records.
Verifies the response to ensure it is accurate and grounded.

Once the agent is deployed, it will operate across all the channels specified in its description, without needing to repeat the setup process for each channel.

Omni Supervisor

You might already know Omni Supervisor as a Service Cloud feature designed for managers to monitor customer service teams. Now, it’s being adapted for agents as well. With Omni Supervisor, you can view overall trends and track agents’ real-time activity. Additionally, you can listen in on active or recently completed conversations, which are summarized in a dedicated column within the table.

Our AI Experts Can Help You Get Started

With all the buzz from Dreamforce, it’s no wonder that many organizations are eager to dive into Agentforce. The great news is that there are immediate benefits to be gained from these tools.

Salesforce Agentforce is the Next Evolution in AI-Powered Customer Engagement. Interested in learning more about Agentforce or need help getting started? We’re here to assist. At Rialtes, we specialize in AI and Analytics solutions for CRM, and we’ve helped clients achieve significant productivity improvements by leveraging AI tools to streamline their business processes.