Navigating Salesforce Agentforce: Structure and Key Capabilities

What is Salesforce Agentforce?

You’ve heard the term all over the place, and everyone’s telling you to look into AI. In fact, by 2025 over 75% of businesses are anticipated to leverage Generative AI in some way. So how do you take advantage? Agentforce is Salesforce’s new hybrid autonomous agent application that is intended to bring generative AI capabilities to businesses and harness those capabilities with business specific data both in a CRM and across a business’s enterprise architecture.  As opposed to other available AI tools and LLMs (Large Language Models) today, Salesforce Agentforce is designed to provide grounded AI responses to natural language prompts that leverages a business’s hard earned, and highly business-specific data all while being straightforward enough to configure with a low-code or no-code approach.  

Salesforce Agentforce relies on the newly released Atlas Reasoning Engine, which is an upgrade from the previously released Einstein Copilot by going from a Chain of Thought reasoning approach to a Reasoning and Acting approach.  Agentforce’s Atlas Reasoning Engine uses Orchestration based on ReAct Prompting, Topic Classification, LLMs for Responses, and Thoughts/Reasoning to provide a conversational experience grounded in a business’s data, that can be fine-tuned through no-code administrators or low-code development.

Besides using a state-of-the-art reasoning engine, Agentforce provides three other critical advantages compared to many other AI agent/ copilot tools hitting the market.  First is the security provided for your business and customers by the Einstein Trust layer.  Second is the ability to ground your AI across the entire suite of Salesforce products.  And third is Salesforce’s customizable agent approach, allowing you to use low-code or no-code approaches to applying tailored AI agents across your business.

The Einstein Trust Layer

The Einstein Trust Layer ensures that your business’s data is tightly secured and protected through both the prompt and response phases of any conversation with an LLM.  Starting with a prompt to an Agentforce agent, Salesforce is able to ensure that your requests are grounded in the appropriate data points to ensure that only relevant and applicable information is sent to the LLM.  Next, before being sent to an LLM, your sensitive data is masked with placeholder text so that no sensitive information will be sent outside of your CRM.  As a final step before being sent to an LLM, your prompt will go through a Prompt Defense layer which applies system policies to include with your prompt to ensure that the LLM behaves in compliance with your system preferences.  Once these first three steps are completed, your prompt is sent to the LLM of your choice. Once the LLM generates a response, any data sent to the LLM is not retained through Salesforce’s Zero Data Retention policy.  This ensures that no third parties can collect or retain you or your customers’ data.  After the response returns to Salesforce, it is filtered through Toxicity detection, Data Demasking, and Audit Trail & Feedback records are created.  These steps ensure that no harmful content or responses are presented back to a user, and that an Audit Trail of all the steps the reasoning engine took are logged in a transparent and accessible audit trail for future reference.

The Einstein Trust Layer protects your data by sending it through multiple layers of transformation, masking, and detection, including a Zero-Retention Policy with the LLM Gateway.

The Einstein Trust Layer is potentially a critical advantage for many businesses over other available copilot tools, as well as potential homegrown options.  By providing a thorough and comprehensive security layer, Salesforce has laid the groundwork to help businesses maintain peace of mind that their business and customer data are safe and secure.

Grounded Data and RAG

Generative AI itself is a great tool, but unless it is tied to data that your business has ownership or access to, it is going to be difficult to drive business value.  In comes Agentforce with Grounded Data and RAG (Retrieval Augmented Generation).  Grounding refers to tying or inputting data from your enterprise systems into an Agentforce conversation or prompt through the use of prompt templates or customized prompts.  By leveraging the data within your Salesforce CRM or any system across your enterprise architecture (see Data Cloud below), you truly unlock the power of Agentforce by combining your data with your AI.

Salesforce’s Customizable Agent Approach

The core approach of Salesforce Agentforce revolves around being able to leverage multiple role specific agents across your business.  These agents can be created from either out of the box templated agents, custom agents that can be built specific to your business use case, or from pre-built agents created by Agentforce partners available on the Salesforce Appexchange.  This unique approach could give Salesforce Agentforce a significant competitive edge, as it will provide businesses with an accessible and flexible model to apply AI across their business.

The initial Out-of-the-Box Agents with the initial October 2024 release of the Agentforce product include Service Agent, Sales Development Rep, Sales Coach, Dispatcher Agent,  and Merchant (covering across each of the four core Salesforce clouds: Sales Cloud, Service Cloud, Marketing Cloud, and Commerce Cloud).  Also available with the initial release of Agentforce is Agent Builder, which allows businesses to create and customize agents through topics and instructions, or through real-time web search.  Subsequent releases for the product have been advertised to include Personal Shopper (planned February 2025), Buyer (February 2025), Scheduling Agent (February 2025) and Employee Service Agent (February 2025).

Agentforce and Data Cloud – A Symbiotic Relationship

Agentforce can not only be tied in with your Salesforce CRM, but it can leverage and be grounded in data across your entire enterprise architecture, from your CRM to your ERP.  This allows you to harness the power of all your relevant data with the generative capabilities of AI to drive value for your business.   Salesforce’s release of Data Cloud allows companies to connect and ground Salesforce AI features with other systems across an enterprise architecture and pulling into a single enterprise data model.  This opens up the possibility to unlock the power not just of your CRM data, but your entire set of enterprise data.

As a “System of Reference” data cloud can be used to create “unified profiles” of any key subset of data across your enterprise architecture.  Need to reference account data across multiple systems in your AI prompts?  Use Agentforce and Data Cloud together to ground your prompts in data pulled from systems across your enterprise architecture to ensure you are providing your users access to all the information they need to do their jobs effectively and efficiently.

A Disruptive Tool to Drive Business Value

Salesforce Agentforce provides some critical advantages over other current competitors in the market today.  Through the Einstein Trust Layer, grounding & RAG, a customizable Agent approach, and access through Data Cloud to data across your entire enterprise architecture Salesforce Agentforce has multiple unique capabilities to help you scale and leverage AI across your business to drive business value.

Interested in learning more about Salesforce Agentforce?  Do you think Salesforce Agentforce could be a good fit for your business or do you want to learn more about it?  Contact us to set up a complimentary AI workshop for your business.  Our complimentary workshop will help educate you and your team on Salesforce Agentforce, as well as identify specific business use cases to focus on to drive business value for your business.  As a leading Salesforce Agentforce and AI implementation partner, we want to help you and your team begin your AI journey on the right foot.