Prior to Dreamforce, we had the opportunity to attend a pre-briefing from Salesforce. In it, CEO Marc Benioff made big claims. Introducing Agentforce, he talked about a new wave of AI and saw AI agents as the most important thing Salesforce has ever done. Agentforce should be available to everyone, not just those who can afford it. “Customers won’t believe what we can offer now,” Benioff said. “No one else can deliver this from a platform.”
Salesforce CEO and founder Marc Benioff acts like the company has introduced something revolutionary. The way he presents Agentforce is over the top. We are used to something from Salesforce, but not in this category.
The bar is set extremely high
In doing so, Salesforce is setting the bar extremely high. First of all, because with such a presentation, there must now be a product that will completely change organizations worldwide. An AI that is more genius than any AI we have seen to date. You can count on us to get to the bottom of that. On the other hand, if Salesforce has developed something revolutionary, the bar will suddenly be raised significantly for the competition.
What is Agentforce?
With Agentforce, Salesforce allows customers to develop their own AI agents easily. Users can call and chat with the agents. Unlike all previous AI agents, these are so intelligent that they can replace employees. Not only do they understand what the customer says or writes, but they can also act so that the customer is helped. All without the intervention of a real employee. Of course, within the set policy. The new AI agents are so intelligent that if an organization adds actions to an AI agent, it will find its combinations on how to apply it.
We were shown an example of a Salesforce manager who had received a sweater that was too large. He called the Agentforce agent to ask if he could get a smaller one. Salesforce had already applied number recognition. The agent immediately knew who was calling and verified that it was the sweater from the last order. Then, the AI agent came up with a proposal to ship a new sweater in a smaller size within five business days. The Salesforce manager indicated that he needed the sweater sooner. The AI agent could not initially help with that.
In this example, the Salesforce manager added the action of having the AI agent view inventory by store. When he called again, the AI agent suggested picking up the sweater today at a nearby store.
Adding low-code/no-code actions
These actions are added to an AI agent through a low-code/no-code editor. This is undoubtedly based on MuleSoft technology, something Salesforce has been making good use of for years in its workflow builder. With a few clicks, a Salesforce administrator can create an action that makes additional data available to the AI agent or a link to an external system to perform tasks. Furthermore, a little context needs to be provided as to what the action is for, after which the AI agent can start working on it immediately.
Voice and chat
These AI agents are immediately available via voice and chat. These are also the best methods for working with them. This way, extra questions can be asked so the AI agent can be sure it is working correctly. With an e-mail or ticket, this is more difficult. For example, an extra question might be: “Did you mean the blue sweater from brand X from your last order on Sept. 1?”
Replacing employees
Based on Salesforce’s demos, this is not yet a beta product but is immediately deployable. Salesforce even states that you can easily scale up your employees with AI agents, especially in a call center for support. Indirectly, that also means that organizations can replace some of their employees with Agentforce. First—and second-line support can be completely automated; organizations only need third-line support.
Already 1,000 agents active
According to Salesforce, over 1,000 agents are already active in organizations. The company expects thousands more to be added after Dreamforce this week. Salesforce is introducing Agentforce templates with AI agents that can be used immediately and customized to fit an organization better.
Salesforce has also established an Agentforce Partner Network through the AppExchange, where more than 20 third-party agents are already available. AWS, IBM, Google, and Workday, among others, already have Agentforce agents ready to go.
Atlas Reasoning Engine
The brains behind Agentforce are the Atlas Reasoning Engine, according to Salesforce. From six individual actions of an AI agent, the Atlas Reasoning Engine can provide a full experience with a variety of use cases in which those actions can be used. It can refine the data links and queries toward an LLM to determine the best answer or action.
If that’s not enough, Salesforce administrators can even help by optimizing queries toward an AI in the prompt builder to better match the desired response to the action it has in mind.
Salesforce not first
While Salesforce would love to claim that it is the first with AI agents from a platform, that is untrue. ServiceNow presented something similar last week but is not yet as far along as Salesforce regarding availability. So, it may be the first to get it working correctly from a platform.
One thing is sure: the AI race is not over yet. The question remains whether Salesforce can make it all happen and whether customers dare to implement it. After all, AI adoption in large organizations still lags behind analyst expectations and predictions. Possibly, this will give another boost.
Also read: Salesforce brings Sales, Service, Marketing and Commerce together in Foundations portal