Salesforce's Agentforce: What You Need to Know About the Autonomous AI Agent Platform

Welcome to another special edition of "The AI Economy." Over the next few days, I'll devote this newsletter to Salesforce's annual customer conference. Today's issue provides a closer look at the company's autonomous AI agent platform, Agentforce, which will be formally introduced at Dreamforce next week.

Calling it "what AI was meant to be," chief executive Marc Benioff and his team are now ready to detail what customers will get from the company's signature AI offering, which it's betting big on. "This is the biggest, most exciting piece of technology that we've been working on," the charismatic executive says, adding that Agentforce will be the only thing discussed at the conference. In fact, Benioff pledges to get at least 1,000 working customers running next-generation agents on its new platform.

Disclosure: I will attend Dreamforce as a Salesforce guest, and the company will cover all my expenses. Salesforce in no way dictated the content of this post. These are my words.

The Prompt

"We're about to enter this third wave of AI," Benioff says. "And this third wave of AI, we're going to know it as agents, and it's Salesforce. We're going to know it as Agentforce. And yes, we're very excited about the fourth wave. We can kind of start to see it, the wave of droids that's coming. Maybe this mythical fifth wave of [artificial general intelligence] will come one day as well. But right now, in the here and now, in the current, present moment, what we're excited about is agents, and this idea of agents is inspiring us back to our core vision." That vision is "to help our customers connect with their customers in a whole new way."

During a press conference on Thursday, he recounted a conversation he had with the CEO of a large banking company in which this person reportedly wasn't able to "explain one piece of value that he has been able to achieve through AI and the huge investment that he was making." It's a common refrain and a sentiment Salesforce wants to change.

Benioff says Agentforce is a platform approach that will "fundamentally transform their business. And they're not going to have to be an expert in a [large language model], they're not going to have to understand [Retrieval-Augmented Generation], and they're not going to have to understand all the key capabilities that you need to know to be a computer scientist—the vast majority of our customers are not."

He pulled no punches when talking about Microsoft and the company's AI Copilot offering, suggesting that customers who implement them have failed to generate value from them. "They really permeated the trust of our customers. They found data leaking. They found the technology was not ready for prime time," Benioff claims.

"This is a moment where technology can really step in. This is the moment where we could ask really meaningful, powerful questions like 'What if these workforces had no limits at all?' That is a moment you're going to see unfold on Tuesday." Salesforce's CEO promotes. "I think it's that really powerful thought that we're ready to deliver a platform—not an LLM, RAG, technique, [or] data set. None of those things. And then it's all of those things because, at the fundamental level, it is time that we're going to be able to demonstrate with a lot of clarity that humans and AI, they're going to drive this customer success together in a trusted way. It's going to be easy for them. It's going to be very low cost. It's going to be low-code, and there's going to be no-code. But mostly, it's going to be a platform. And it will be a platform that they already have, already [are] familiar with. There's nothing new to install."

Let's Dive Into Agentforce

First, Agentforce is not a single application but a suite of AI-driven agents built to help workers with service, sales, marketing and commerce tasks. The platform isn't designed to automate everything in the workplace, though that makes sense—Salesforce's AI is best optimized for tackling what's in its proverbial wheelhouse.

Four components make up Agentforce, which we'll explore later: Several Salesforce platform applications are at its core. The collection includes a host of pre-built agents and mechanisms to do things your way and a sizeable partner ecosystem.

In a way, Salesforce believes organizations building out their AI agent strategy is akin to the state of self-driving cars:

"In the autonomous driving world, you have a range of assistance. At level zero, for example, you might have zero assistance. Level one might give you light assistance. And at level five, you might have full assistance or autonomy. You can think of autonomous AI for the enterprise in much the same way," Clara Shih, Salesforce's AI CEO, says in a statement.

And like a self-driving car, the company claims its new platform uses real-time data to adapt to "changing conditions and operates independently within an organization's customized guardrails."

When Agentforce makes its official debut Tuesday at Dreamforce, it will already be used by a handful of customer companies, including OpenTable, Saks, and Wiley.

Its release hasn't been a secret—Benioff revealed the news following the company's fiscal year second-quarter earnings. But this is the first time we get a detailed breakdown of what this platform can do. It also provides a 50,000-foot view of how Salesforce is combining pieces to help customers embrace AI and be more productive in the customer relationship sector.

Agentforce Safety and Guardrails

As with any AI development, there is skepticism that tech providers have installed the necessary protocols to ensure the AI won't return hallucinations or incorrect information. It's one of the most prominent fears organizations have that makes them reluctant to embrace the technology. However, Salesforce is touting that Agentforce has a strong trust layer that'll minimize any cause for concern. Shih describes it as "battle-tested."

When asked about the company's confidence in letting Agentforce agents operate independently with designated guardrails, Silvio Savarese, Salesforce's Chief Scientist, says that the trust layer implemented will "block situations that are undesirable." He further added that there's a component with the ability to "actually provide transparency in terms of what decisions are made. So, it's possible to let the agent explain why certain decisions have been made. And this allows [us] to open the hood [to get] clarity on certain steps. There is also going to be an opportunity for adding confidence values, so much [confidence is there in] an action being made. This allows you to escalate to the human supervisor if certain tasks are being assessed with low confidence."

This point is one reason Shih believes Salesforce is winning customers over with Agentforce: "Because you need to be able to seamlessly hand off to a human whenever the confidence score drops. And what we're seeing very conservative regulated industries do is they'll often start with our existing human-in-the-loop use cases. Once they see good results from that, those workflows will graduate to the customer self-service agent."

The Four Components of Agentforce

Salesforce Platform

What would Agentforce be without some tie-in with its sibling applications? Perhaps the most important player here is Salesforce's Data Cloud. The product formerly known as Salesforce Customer 360 and Genie launched in 2020 and now powers the company's Einstein AI. As the repository for customer data and metadata across systems, it's a critical element of Agentforce's performance and success.

The data from Data Cloud can be used to train the AI agents on the customer's needs. Because it's proprietary data, RAG will help reduce hallucinations and enable the bot to provide quality responses.

Now that AI agents have a library of data to consult, they need a brain to process that information. This is where Salesforce's new Atlas Reasoning Engine comes in. Developed in-house, the company claims it is built to "simulate how humans think and plan." Incorporating reasoning into models is growing in popularity. Salesforce isn't the only one working on this. On Thursday, OpenAI released o1 (codenamed "Strawberry"), its first model capable of handling complex queries.

Nevertheless, the Atlas Reasoning Engine will evaluate user prompts, refining them for clarity and relevance before searching for the most relevant data. Then, it will establish an execution plan. After that, the process begins anew, refining once more to ensure accuracy and relevancy and that answers are grounded in trusted data.

However, it's worth noting that the Atlas Reasoning Engine will not be available at launch. It's scheduled to be released in February 2025.

After acquiring MuleSoft in 2018 for $6.5 billion, Salesforce is making it an integral part of the Agentforce platform, leveraging its Salesforce Flow, MuleSoft and Apex methods. This makes sense as it enables customers to connect all their information spread throughout public and private clouds and data sources. Organizations will not have all their data confined into Salesforce's CRM, no matter how badly Benioff might want it. Using a middleware service like MuleSoft opens up integrations with any company's workflows or systems that may have once been considered incompatible with Salesforce.

Out-of-the-Box Agents

So you've unboxed Agentforce and are ready to start with AI but don't know what you want to build? Salesforce has a solution for you: Pre-built bots that address multiple needs. The company claims these off-the-shelf options are easy to customize and don't require coding to deploy. They can be set up "in minutes" and work across all channels.

There are eight of these AI agents available at launch:

• Service Agent: An AI-powered chatbot to handle a wide range of service issues without needing preprogrammed scenarios
• Sales Development Representative: An AI agent that will interact with prospective customers around the clock, answering questions, troubleshooting and scheduling meetings with human agents based on CRM and external data
• Sales Coach: A personalized assistant for sales teams, offering role-play sessions and leveraging Salesforce data and generative AI to train representatives on pitches and objections
• Merchant: An AI agent made for e-commerce, assisting merchants with setting up a site, goals, personalized promotions, and product descriptions, along with providing data-driven insights
• Buyer: A B2B-centric agent that helps buyers find products, make purchases and track orders through the use of chat or sales portals
• Personal Shopper: The digital concierge on an e-commerce site or messaging app, providing users with personalized product recommendations and assisting with search queries
• Campaign Optimizer: An AI agent that will automate an entire campaign lifecycle, analyzing, generating, personalizing, and optimizing marketing campaigns based on an organization's business goals
• Agentforce: Formerly Einstein Copilot that launched in February 2024, this conversational AI assistant for CRM has been rebranded and upgraded. It can now retrieve data, reason, build a plan and take action. It's an embeddable agent constructed to support an employee's workflow.

Build and Deploy

One size won't necessarily fit all. Should one of the aforementioned AI agents not work for an organization, Agentforce allows the construction of customized and unique agents without extensive technical know-how. The platform will enable the development of agents, models or even prompts, all through low code.

• Agent Builder: Customize out-of-the-box agents or build new ones for any use case using Salesforce Flows, Prompts, Apex and MuleSoft APIs. Define the topic, compose natural language instructions on it, and create a library of actions for it to choose from.
• Model Builder: Register, test and activate custom AI models and LLMs across Salesforce using this tool. Bring your own API keys for the LLMs of your choice and test them in a playground before activating them.
• Prompt Builder: Create easily customizable prompt templates using your CRM or Data Cloud data to improve the output of the generated results. Prompt Builder embeds the generative experience in the workflow, whether it's automated, on a Salesforce Lightning record, or in an agent's actions.

Agentforce Partner Network

At launch, Salesforce's Agentforce will be supported by many partners, including Amazon Web Services, Box, Certinia, Copado, Coupa, Google, Honeywell, IBM, Workday, and Zoom. This ecosystem has produced over 20 agents and agent actions that can be found now within Salesforce's AppExchange marketplace.

This familiar development playbook makes Salesforce appear like a good corporate citizen. It recognizes that it's not hosting all the data in the world and isn't the best in every sector and industry. So, with its relationship with third-party providers, it's helping its users eliminate stumbling blocks to derive greater value out of their data and build better relationships with their own customers.

Pricing

There have been concerns about how much Agentforce would cost users, and some are prodding Salesforce to provide pricing clarity. The company says that its AI agent platform starts at $2 per conversation, with standard volume discounts. Benioff claims the cost savings as a result of Agentforce will be "incredible," and the company calls the pricing a "fair economic deal," believing that AI can be the level 1 tier support, which is a more expensive option, allowing those in level 2 to focus on more significant issues.

Benioff offered this endorsement, "Mark my words...this is the direction that all companies will go because this is the only direction that will work to help customers achieve success, which is our most important value."

Agentforce for Service and Sales will be generally available on October 25, 2024.

Editor's Note: Subscribe to "The AI Economy" today to receive additional analysis from Dreamforce, plus a weekly roundup of AI news you may have missed.

End Output

Thanks for reading. Be sure to subscribe so you don’t miss any future issues of this newsletter.

Did you miss any AI articles this week? Fret not; I’m curating the big stories in my Flipboard Magazine, “The AI Economy.”

Connect with me on LinkedIn and check out my blog to read more insights and thoughts on business and technology.

Do you have a story you think would be a great fit for “The AI Economy”? Awesome! Shoot me a message – I’m all ears!

Until next time, stay curious!