Salesforce’s Agentforce is a suite of integrated tools for creating autonomous AI agents. Agents use powerful AI, data, and integration to make informed decisions and perform tasks independently. Out of the box, users can assign prebuilt agents or customize their own for roles that would previously be the domain of humans.
Imagine a world where computers interact with customers and autonomously make informed decisions consistent with your company’s perspectives and objectives. Salesforce wants to make this a reality with its new Agentforce platform.
Agentforce promises to augment each worker’s ability to effectively connect with and serve customers across multiple engagement channels — providing meaningful, actionable, and tailored customer support around the clock. There are also internal use cases, such as on-demand job coaching for staff.
Will Salesforce deliver on its big promise?
“We will all wait with bated breath on whether Salesforce has cracked the code on artificial intelligence (AI). But putting sales agents to the side, whoever cracks customer service AI first will have a huge head start on the windfall of AI spending that will occur in the very near future.”
Ben McCarthy, “Salesforce Agentforce: All Hype, or Will it Make History?”, SFBen, September 20, 2024
Either way, it’s time to explore the new platform everybody’s talking about. By the end of this article, you should have a basic understanding of Agentforce, what it does, and how to prepare for the future of AI agents.
What is Agentforce?
Salesforce’s Agentforce platform is a suite of integrated tools that allow people and teams to create customized “agents”—or groups of agents—capable of accomplishing routine and complex tasks faster, more efficiently, and more effectively.
Agentforce agents autonomously and proactively make decisions and take actions using large language model (LLM) data analysis while abiding by brand—and objective-based parameters.
These agents, capable of working 24/7 across multiple channels, can also identify circumstances that require human intervention and escalate matters accordingly.
At its core, Agentforce’s primary objective is to create task- and industry-specific agents capable of performing tasks that were previously the domain of humans. Agents can be designed to service internal and external constituents, but they are commonly positioned in customer-facing roles. The applications are, in theory, unlimited.
So, to keep things simple, we’ll focus on their ability to provide relevant, accurate, and genuinely helpful service to customers.
We can think of agents as chatbots on steroids. When well designed — and fueled by high-quality data — Agentforce agents (and agents in general) may significantly leverage the ability of enterprises and customer-facing enterprise employees to deliver better customer experiences guided by insights gleaned through the analysis of vast data repositories.
And because those interactions will be guided by clearly established brand and content guardrails, agents will still provide the means to trigger human interaction when and where it’s necessary to achieve desired outcomes.
For example, a financial services company might wish to deploy agents in its self-service portals and other messaging channels. Unlike chatbots, agents can access APIs outside of Salesforce to communicate with customers, make informed decisions, and complete various approved tasks. The agent’s role and associated guardrails mean it can autonomously decide to schedule a meeting, but it can’t do more than it’s allowed.
What is Salesforce’s Goal for Agentforce?
From Marc Benioff’s perspective, Agentforce is ushering in an “unlimited age” that will redefine how humans work. He and others refer to this upcoming phase of technological development as the “Agentic Era.” With the promise of unlocking capacity across industries, Agentforce’s ambitions are for digital labour to lower the cost of scaling operations.
Imagine a fleet of customer service agents available 24/7, with data processing and AI-reasoning capabilities. This experience would outperform basic chatbots on so many levels. And any customer who’s engaged a chatbot can attest that this is a laudable goal.
A key part of this larger goal is to make Agentforce easy for everybody with powerful natural language processing (NLP). Workers can design agents using templates or natural language prompts, making it possible to develop and refine tailored agents without specialized expertise. Customers can also interact easily with agents using natural language.
How Much Does Agentforce Cost?
Agentforce pricing is currently based on conversations — or interactions between agents and their intended audience. Salesforce has established baseline rates of $2 per conversation, with lower per-conversation costs based on interaction volume. Visit their official pricing page to learn more.
“Agentforce for Service and Sales will be generally available on October 25, 2024. Some components of the Atlas Reasoning Engine launch in February 2025. Agentforce pricing starts at $2 per conversation; standard volume discounts apply.”
“Salesforce Unveils Agentforce: What AI Was Meant to Be“, Salesforce.Com News & Insights, September 12, 2024
Of course, whether Agentforce’s “conversations” are cost-efficient will depend on the nature and volume of interactions they replace and whether those conversations produce outcome success rates equal to or higher than they would if carried out by human sales and service agents.
This is very much the early stages of AI agents, and much remains to be seen regarding pricing and competition.
How Does Agentforce Work?
Agentforce is powered by the Atlas Reasoning Engine, a sophisticated, AI-driven reasoning powerplant that does the heavy lifting when autonomous, automated agents meet end users looking for answers. These end users can be employees looking for job coaching, online shoppers tracking their shipments, and so on.
Agents can interact through text and voice – even answering the phone, as we saw during the live Agentforce demo at Dreamforce 2024. From the end user’s perspective, things may not look or sound different. Once a user enters their query, the Agentforce agent seamlessly, quietly, and efficiently goes to work. That’s where Atlas earns its keep.
First, Atlas evaluates the prompt against the agent’s clearly defined goal to create a strategic plan of attack while enriching the prompt to include intent data alongside the keyword data.
Next, Atlas initiates a comprehensive data search across its designated environments. This search could include structured data from Salesforce Sales Cloud, Service Cloud, and more — including unstructured data sources like emails and call logs. IDC estimates that “unstructured” data sources may represent as much as 90% of enterprise-held customer data.
Finally, Atlas triggers Data Cloud’s Vector Database to simultaneously analyze relevant data from all sources to formulate a targeted response. The agent then delivers that response to the end user through the same communication channel where the initial query was made.
Almost instantaneously, the agent naturally responds to the end-user in a human-sounding cadence. This interaction feels smart because Atlas processes LLM and CRM data to “retrieve, plan, evaluate, and refine” its responses.
How Capable is Agentforce Now?
Because the technology is quite new, Agentforce’s capabilities are somewhat aspirational. The Salesforce keynote, while full of excitement, addressed this at times.
“…it’s a little bit of a Highwire act for a technology company because we’re doing something that no one has ever done before”.
Marc Benioff, Dreamforce 2024 Main Keynote with Marc Benioff | Welcome to Agentforce | Salesforce, Salesforce, September 17, 2024
Today, Agentforce provides several “out-of-the-box agents” that require no code and only minutes to get up and running. They include:
- Service Agent
- Sales Development Representative
- Sales Coach
- Merchant
- Buyer
- Personal Shopper
- Campaign Optimizer
While Agentforce agents are designed to perform effectively immediately, they also learn and improve. As they experience human interaction and dynamic situations, agents “teach” themselves with real-time feedback — elevating their long-term potential beyond that of simple chatbots with rigidly established scripts.
How Are Agents Created?
In contrast to applications that require significant IT support and end-user training, Agentforce’s powerful technology lets you hit the ground running.
Agentforce’s Agent Builder helps you build agents using out-of-the-box templates, natural language input, and low-code. By design, it’s easy to establish guardrails and parameters that protect your business and customers from unapproved experiences.
Agentforce’s Agent Builder walks users through a process to establish critical operational parameters, including:
- Agent Role: This definition establishes which topics the agent is empowered to address.
- Data Source: Agents are structured to access specific data sources. Those sources can be structured, like Salesforce data — or unstructured, like raw email or voice memos. Agents can also draw data from other LLM resources designated when the agent is created.
- Topic Actions: Agents can only process actions defined during their creation. This ensures that agents don’t produce results beyond their intended scope.
- Action Guardrails: Just as agents are not authorized to exceed the scope of their designated topics, they can also be fine-tuned to withhold results within those topics.
- User Channel: Agents are designed to deliver results through those defined channels used by their designated internal and/or external audiences.
Is Agentforce Just a Rebranding of Einstein Copilot?
TLDR: No, it’s not.
When Salesforce released Agentforce in October of 2024, it took care to distinguish its capabilities from Einstein Copilot AI. Basically, Einstein Copilot was like a building block for Agentforce’s significantly more robust capabilities.
Einstein Copilot is powered by the same Retrieval Augmented Generation (RAG) capabilities that drive Agentforce’s Atlas Reasoning Engine. But Agentforce’s key differentiators come down to:
- Autonomous operation: Where Einstein Copilot required direct human input to trigger each activity, Agentforce can take human prompts and execute related tasks autonomously.
- Complex task processing: Agentforce’s ability to perform multiple, complex tasks far outpaces Copilot’s simple task execution capabilities.
- Data use: Agentforce utilizes RAG and Vector Database capabilities to process structured and unstructured data in real-time — an enormous advance over the script-based processing available to Copilot.
- Integration potential: Agentforce is designed for deep integration with Salesforce tools, third-party applications, and databases. Copilot’s integration was basic, limited mainly to the confines of the Salesforce ecosystem.
- Customizability: Where Copilot’s customization potential was limited, Agentforce is designed for in-depth customization.
“One of the core strengths of the Atlas Reasoning Engine is its contextual understanding. Traditional support systems often rely on rigid workflows and keyword-based searches. However, Atlas leverages NLP and deep learning to comprehend the context behind customer inquiries. This means it doesn’t just surface relevant knowledge base articles but also connects the dots between customer history, case specifics, and even sentiment analysis, delivering recommendations that are far more aligned with the customer’s needs.”
The Brain Behind the Agents: Unveiling the Atlas Reasoning Engine Behind Agentforce, Rialtes, October 21, 2024
Unlike Einstein and its peers, Atlas operates on natural language processing to consider the content of knowledge-based articles in the context of customer histories, situational specifics, and even emotional analysis to provide responses that more effectively answer customer needs. Atlas doesn’t just point to an answer that surfaced in another article — it produces a detailed, tailored response that directly addresses the customer’s particular inquiry.
That response is formulated through Atlas’s use of:
- Solution-Based Reasoning. Whether supporting live agents or interacting directly with customers, Atlas offers case history-based solutions in real-time, suggesting the next best actions to resolve challenges more quickly, accurately, and effectively.
- AI-Fueled Insight. Atlas leverages AI to identify recurring problems and complaints and applies those insights to quickly and effectively resolve core issues. It can also utilize those insights to suggest actions to prevent future issues.
- Feedback-Fueled Situation Management. Unlike the more linear, algorithm-driven processes of technologies like Einstein Copilot, Atlas adapts its efforts according to new inputs and changes course to provide timely, complete, and thoughtful responses.
- Multi-Platform Integration. Working within predetermined parameters, Atlas can tap multiple data sources, including Sales Cloud, Service Cloud, and designated external platforms, to produce responses that factor in comprehensive customer histories.
Is Agentforce Right for Your Business?
While Agentforce’s current capabilities are impressive, it’s not the only AI agent game in town. As this sentence is being typed, G2 lists 128 entries in its “Best AI Agents” category, and Agentforce isn’t yet among them.
“…in terms of comparing a new shiny object to something we already have, it’s a lot easier to look for the new stuff and say …This will solve all of our problems. If we just get this one tool approved, then everything will be better. And it’s rarely the case that happens. …historically, we’ve been much more focused on bringing in new functionality.”
Victoria Savio, “Innovating Asset Management with Victoria Savio, Senior Vice President and Head of Business & Data Enablement at Voya Investment Management,” Riva, August 7, 2024
From Salesforce’s perspective, many industries are ripe for applying Agentforce’s capabilities. The list will likely grow as Agentforce technology matures and its applications become more apparent and widely understood. As Salesforce puts it, that list includes:
- Healthcare
- Banking
- Retail
- Operations
- Customer Experience
- Analytics
- Information Technology
- Finance
But will your business benefit from AI agent technology? Assuming you have customer-facing teams or web-based customer service technologies, Agentforce’s powerful capabilities could be beneficial— and possibly even game-changing. The following self-reflection questions can help start your evaluation process:
- Does your business currently use Salesforce Sales Cloud? Because Agentforce requires a Sales Cloud license, this is a threshold determination.
- Is your business customer service-intensive? While AI agents can be designed to support internal audiences, Agentforce has a strong CX focus that may be a critical consideration.
- Do you currently utilize AI chatbot or copilot technology? If you’re new to AI technology, starting with Agentforce is like jumping into the deep end. However, for those with experience, Agentforce can offer significant performance improvements, potentially sweetening the appeal of integration.
- Do you practice good data hygiene to ensure high-quality CRM data? This is a critical question because, like any AI technology, Agentforce’s performance depends on good data. Poor CRM data can lead to poor customer experiences.
- Does your business need a tool like this? Avoid ‘shiny object syndrome.’ Perhaps you’re philosophically ready to embrace Agentforce, but that doesn’t mean you have a significant use case for it.
So, is Agentforce right for your business? There’s no short answer. However, given its potential to significantly enrich employee and customer experience, it’s worth considering.
Reliable Data Comes First
Agentforce could be a meaningful enhancement to Salesforce’s arsenal based on an early, high-level look at its capabilities. But it’s only as good as the data it relies on. That’s why enterprises must do the hard work of data quality improvement before investing in a new technology that won’t measure up if the data quality foundations necessary for success aren’t in place.
Riva powers Salesforce with clean, accurate data, enabling Agentforce insights that deliver real business value. By automating data flow from sources like Outlook, Riva eliminates inefficiencies and ensures your customer data strategy drives success.
We encourage you to contact us for more information — so your business will be ready to reap the rewards of technologies like Agentforce.