Salesforce: The Path to 1 Billion AI Agents
It takes only minutes to create the autonomous digital agents, but weeks or months to bring them to full speed.
Welcome to the Cloud Database Report. I’m John Foley, a long-time tech journalist, including 18 years at InformationWeek, who then worked in strategic comms at Oracle, IBM, and MongoDB. I invite you to subscribe, share, comment, and connect with me on LinkedIn.
I attended Salesforce’s recent Agentforce event in New York City with one question in mind: How much progress is being made toward the 1 billion AI agents that Marc Benioff boldly set as a goal at Dreamforce 2024 in San Francisco?
Dreamforce 2024 was just nine weeks ago, in mid-September, so this may seem an unfair question. However, Benioff’s ambitious goal is to hit 1 billion agents by the end of 2025. And the market — including competitors such as Microsoft, Oracle, ServiceNow, Workday, and SAP — is moving fast.
Salesforce had some updates to share at Agentforce NYC, which took place November 20 at Javits Center. First, and most important, the company announced Agentforce Test Center, where developers and others creating agents can test the ways a customer might pose questions and interact with a digital agent. The Test Center can generate hundreds of synthetic interactions and gauge the results, so agents can be fine-tuned for deployment at scale.
Related to this, Salesforce introduced “sand box” development testbeds for both agents and Data Cloud, its fast-growing data lake service. In earlier posts, I referred to Data Cloud as the superpower behind Agentforce. There are also new monitoring, observability, and usage capabilities for AI agents — the kind of enterprise tools that are needed to deploy millions and millions of these digital, autonomous personas.
For a refresher on Agentforce, see my earlier post below from the launch event, which includes an architectural diagram showing how the pieces fit together.
Like onboarding an employee
“Build an agent in 10 minutes.” Thus touted the digital signage around Javits Center. Sounds easy — too damn easy to those of us who have been reporting on new technologies as long as I have.
I pressed the Salesforce team and customers to better understand what ramp-up really looks like. Some of the people I heard from included Rahul Auradkar, EVP of Data Cloud, and Sanjna Parulekar, VP of product marketing. On the customer side, I talked to Stephane Moulec, CTO at Good360, and Greg Beltzer, head of technology with RBC Wealth Management.
As I expected, it may take only minutes to get a Salesforce AI agents up and running, but days, weeks, or months to bring them up to full speed. RBC built and deployed its first prototype in six weeks. Good360, a non-profit, has been working on its first agent for two months, with plans to go live early in 2025.
The timeline will vary for each organization. My rule of thumb is that launching an AI agent is akin to onboarding a new employee. It can be fast and easy to get them started, but there’s a learning curve before they’re truly expert at the job.
By all accounts, the process of creating an Agentforce agent — using the Agent Builder drop-down menus to define roles and actions, and APIs to connect to data sources — is the easy part. Team adaptation is more complicated. By that I mean how employees will use and collaborate with digital agents and, crucially, everything related to the customer experience.
‘Imagine what we can do’
Early adopters of Agentforce include Adecco Group, Good360, OpenTable, RBC, Saks, and book publisher Wiley.
It will come as little surprise that some businesses are starting out by implementing AI agents for internal use cases — putting on their AI training wheels with employees before customer-facing engagements.
Example #1: Good360 plans to use its first agent to automate some of the processes involved in determining where and how to move the thousands of truckloads of donated products it distributes to people in need. That should let the non-profit’s staff focus more time on community-facing support. “We move $3 billion of products,” Moulec told me. “Imagine what we can do if AI helps us.”
Example #2: RBC built and deployed its first agent to assist its thousands of financial advisors in the U.S. The initial use case is to summarize client info in advance of advisory calls. Users rated the quality of these summarizations 4.8 out of 5.0 — not bad, and presumably that will only get better. In terms of productivity, the agents are saving advisors 90 minutes of research work each time. RBC has a half-dozen additional use cases queued up and many more under consideration.
What about worries that AI agents will displace employees? So far, that hasn’t happened in these two examples. In fact, RBC’s US-based advisors have shown an eagerness to get their hands on AI tools. “Their time is valuable,” Beltzer explains. “Anything I can do to make their time more efficient, they love it.”
Early signs of acceptance
So there are signs of progress. But will this add up to 1 billion Salesforce agents by the end of 2025? Too early to know. It was just 24 days ago that Benioff announced that Agentforce was generally available.
The empirical evidence, based on my observations, suggests strong interest. At Dreamforce nine weeks ago, the goal was to kickstart 1,000 agents at that event. Attendees created 10 times that many. Likewise, the Agentforce Zone in New York was crowded, as you can see from the photo.
There’s another early indicator — one more important than my own show-floor observations. Salesforce released a report indicating that agents and other AI technologies are helping to drive retail sales. This alone could give a boost to acceptance of AI agent technology during the holiday shopping madness that’s now upon us.
Here are a few interesting nuggets from the Salesforce report:
AI contributed to 17% of global purchases over a seven week period
Cyberweek is expected to drive $311 billion in global sales, with AI and agents influencing up to 19% of orders
Digital retailers using GenAI and agents increased their average order by 7%
Retailers using AI-powered agents doubled customer service engagement
I assume there’s not a direct link between these data points and use of Agentforce agents, since the technology is still so new. But it’s the kind of data that should cause retail business leaders to ask, “Is there more we should be doing with AI agents?” As a consumer too often frustrated by poor customer service, I’m eager to see that happen.