This is "The AI Economy," a weekly LinkedIn-first newsletter about AI's influence on business, work, society and tech and written by Ken Yeung. Sign up here.
Get ready for a proliferation of autonomous agents.
Salesforce’s Agentforce platform is now generally available, publicly launching more than a month after its introduction. With it, the company is ushering in the democratization of sophisticated AI bots, providing a solution in which anyone can quickly develop AI agents that connect with their enterprise data and will help solve sales, service, marketing, and customer requests.
The launch is expected to intensify Salesforce’s ongoing battle with Microsoft. CEO Marc Benioff has criticized the rival’s Copilot AI, claiming customers are dissatisfied with the offering. He has even gone so far as to liken it to a modern-day “Clippy,” taking a jab at Microsoft’s AI capabilities.
Although Salesforce isn’t the first to have an AI agent builder, its solution appears unique in requiring no technical knowledge to operate. This eliminates the need for companies to hire specialized developers to develop bots, streamlines the process, and removes an obstacle to innovation.
An ‘Inflection Point’ for Salesforce
Over 25 years ago, Salesforce pioneered cloud computing with a mission to help companies better understand their customers. Back then, the default was organizations applying a “do-it-yourself” mentality to create solutions that worked for them, duct-taping together databases, APIs, their own CRM with contacts and leads, and other applications. Then, Salesforce entered the picture and provided a unified subscription offering.
It’s a pattern the company wants to replicate for artificial intelligence. “We’re at an inflection point with Agentforce,” Salesforce’s Senior Vice President of Product, Salesforce AI Platform, Adam Evans, remarked at a press briefing this week. “18 months ago, we had this really awesome breakthrough come through with large language models. And everyone wants to move into this. And it’s kind of like what that first rush was maybe 25 years ago, [with] some of the same patterns playing out. We have customers spending millions of dollars trying to figure out how to bring in this awesome new technology, which is really going to revolutionize everything about business and how customers interact with brands…and how they bring this into their companies.”
Evans sees Agentforce as the key to bringing autonomous AI agents to the enterprise mainstream, eliminating the need for technical specialists or piecing together multiple services to achieve desired results. With organizations still confused about using AI as part of their strategic workflow, offering a tool that streamlines agent production and distribution could alleviate some concerns, especially from businesses that don’t have the budget to hire developers to build out these chatbots.
Salesforce believes its customers want a better experience. At Dreamforce, Chief Executive Marc Benioff spoke about the “a-ha” moment people received when riding in a Waymo self-driving car for the first time. That feeling is what customers want from using AI, but Salesforce claims they’re not receiving it today—something Evans emphasized:
“Do they want to build a car by taking all the parts and building their own car? Or do they just want to take a Waymo and have their problem solved?”
Salesforce Agents vs. Microsoft Copilots
Before we continue, let’s clarify what we mean by an “agent.” Salesforce defines these bots as “having the ability to search through data, all of the knowledge that’s necessary to complete a job, being able to analyze that data, to come up with a course or plan of action, and, of course, being able to execute that so not just answer questions, but transact, and to do that with proper guardrails and grounding on data.”
They’re different from copilots, at least in Salesforce’s eyes. “Copilots are less sophisticated than agents, primarily because, at the end of the day, copilots are all about bringing back information,” Brent Hayward, the company’s head of competitive intelligence, explains to me. “Where agents take a step change is they reason with that information [and] they use the AI to ultimately call and drive action. And it’s autonomous and semi-autonomous action.”
“Microsoft would like you to think of ‘agent’ as the new synonym for a workflow. And in that context, it’s like calling an action in a chatbot, whether it’s the decision tree that’s been replaced by AI or maybe the ability to do an automated task,” he says. “When you think of agents, think of literally the digital equivalent of a person doing a task or job. Think of, ‘I’m in a call center, and today I am fundamentally constrained by the labor I have, and I’ve had to put chatbots and all these other things; I use words like deflection, which is not the best customer experience because I only have so much labor.’ But what if I could clone 1,000 of me, and what if it was very capable of engaging with you, of reasoning with you, of understanding your needs, but then being able to take action? Action is actually ‘Let me book your service,’ ‘Let me change your delivery,’ and ‘Let me fulfill your prescription.’ Think of ‘agent’ as an agent-centric, new set of workers, not a set of static workflows that get called, and I think with that frame, you’ll see how this is just fundamentally different.”
There’s no love lost between Salesforce and Microsoft regarding AI. In the days leading up to Agentforce’s launch, Microsoft announced that its AI agent builder will be coming to Copilot Studio in a public preview next month. The tool was introduced earlier this month at the company’s Build conference and aims to create more intelligent bots. Not one to pass up a good opportunity, Salesforce quickly responded, with Chief Business Officer Kendall Collins posting to LinkedIn that “Copilot’s a flop because Microsoft lacks the data, metadata, and enterprise security models to create real corporate intelligence. That is why Copilot is inaccurate, spills corporate data, and forces customers to build their own LLMs. Clippy 2.0, anyone?”
Microsoft fired back with its own post, revealing that a year after Microsoft 365 Copilot became generally available, 60 percent of Fortune 500 companies are using the technology.
I contacted Microsoft before Agentforce’s launch for comment but haven’t heard back. If the company responds, I’ll update this post accordingly.
Whether the anti-Copilot marketing will work to Salesforce’s advantage remains to be seen. However, customers appear to be interested, at least initially. At Dreamforce, Benioff aimed to get at least 1,000 customers onboarded to Agentforce—more than 10,000 created their first agent by the end of the conference. Now that the agentic platform is open to all, all eyes will be on Salesforce’s subsequent quarterly earnings to see if this “transformational moment” Benioff realized months ago will pay off for the company.
How Agentforce Works
Agentforce has four central parts: the Salesforce platform, eight ready-to-go agents, building tools for agents, models and prompts, and a budding partner ecosystem.
Salesforce’s Data Cloud, formerly Customer 360 and Genie, is central to the Salesforce Platform. It powers the company’s Einstein AI and is the repository for customer data and metadata across systems. Using Data Cloud as the primary source of truth, Salesforce ensures agent responses will be factual, with minimal hallucinations.
For organizations with information scattered outside the Salesforce platform, Salesforce is leveraging its ZoomIn acquisition to ingest previously siloed data from legacy systems, data warehouses, and lakes.
Agentforce’s intelligence comes from Salesforce’s Atlas Reasoning Engine. Unlike other vendors that might outsource this to OpenAI or another AI model provider, Salesforce developed this capability in-house and claims it can “simulate how humans think and plan.” Though it was introduced as part of the platform, the “brain” will not be available until early next year.
Another critical platform element is MuleSoft, which Salesforce acquired in 2018 for $6.5 billion. Customer organizations can leverage the app to integrate Agentforce into their workflows or systems that may have once been incompatible with the Salesforce platform.
Beyond that, Salesforce has a slate of pre-built AI agents to help customers get started. These include bots for customer service, sales, commerce, and marketing.
If those agents don’t meet customers’ needs, customers can create their own using Agentforce’s builder tools. Salesforce’s Agent Builder, now available, simplifies agent setup and activation with no-code and low-code options. A helpful Wizard guides users in selecting and setting up their agent. They can then specify the job to be done by the agent by defining topics, writing natural language instructions, and creating a library of actions for the agent to choose from.
The last part of Agentforce is the partner network, consisting of companies such as Amazon Web Services, Box, Certinia, Copado, Coupa, Google, Honeywell, IBM, Workday, and Zoom. Some of them have produced their own agents and agent actions and are available within the Salesforce AppExchange marketplace.
The Future of Agentforce
Now that Agentforce has launched, what can we expect in the future? Evans reveals there are “a handful of different kinds of dimensions” Salesforce plans to invest in. One of which is having multiple-interacting agents “moving into longer running, deeper thought, maybe even slower kind of tasks versus systems that are designed around rapid, high-performance responses.”
In addition, he believes there will be more work done around general at-scale automation, thanks to AI allowing tech companies to go after a long tail.
He affirms that whatever Salesforce does next will follow its North Star, helping businesses run and connect with customers. “It’s not about, let’s say creating a copilot or something that’s about general productivity of people like inside a Word processor or something like that, to summarize things, although it can be that, but that is a very myopic part of it. The question we ask, and what we hear from our customers, is: ‘How do we apply this technology to just make our businesses run in a completely different way?’”
Looking ahead, Evans describes this moment as the “first chapter” in a long, exciting, and likely swift innovation cycle aimed at transforming businesses with AI.
Leave a Reply
You must be logged in to post a comment.