Salesforce Agents 2.0 and the Trillion-Dollar Upside in Digital Labor
New Agentforce 2.0 platform integrates with Slack and brings an expanded menu of prebuilt skills.
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. This post is sponsored by Salesforce.
We can all stop wondering what the agentic future might one day look like. It’s here. It’s called digital labor. And it promises to be a huge growth driver for businesses and the global economy.
“It’s not some vision or in-the-future idea. It’s what’s happening right now,” Salesforce CEO Marc Benioff declared at the launch of Agentforce 2.0 in San Francisco on December 17. “We’ve already crossed the bridge to digital labor.”
Agent technology is speeding ahead, even faster than the tech industry’s usual breakneck pace. It was just 12 weeks ago that Salesforce premiered v1 of Agentforce as an out-of-the-box solution for sales, service, marketing, and commerce. The new Agentforce 2.0 now takes the platform further, with Slack and Tableau integrations, an expanded menu of prebuilt skills, enhanced reasoning, and other innovations. It’s also easier to customize these autonomous, digital agents, which is significant because that’s probably what many organizations will choose to do.
Other tech vendors are likewise moving AFAP — as fast as possible. In the past few weeks, Google introduced Gemini 2, which it describes as an AI model for the agentic era; Microsoft previewed new agents in Microsoft 365; and AWS rolled out multi-agent collaboration for its Amazon Bedrock foundation-model service.
Benioff believes Salesforce is in front of these competitors and others, and he’s not letting up. The company just surpassed 1,000 customers who have signed on to Agentforce, and it’s hiring more salespeople to keep that going. “Time for us has definitely accelerated,” Benioff told media, analysts, and customers packed into the conference room at the St. Regis Hotel, across the street from where Dreamforce 2024 took place in mid-September.
Eating the agent dogfood
The business case for the agentic workforce, and how Salesforce talks about it, is changing, too. Agentforce is now billed as a “digital labor platform,” which sounds like dry toast until you think about all of the other life-changing digital technologies that preceded it: digital payments, digital images, digital twins, digital media, etc.
To hear Benioff tell it, digital labor is poised to be the next big, disruptive digital transformation. And I think he’s right.
It’s like a lightbulb went on for Benioff as the implications for driving business productivity — and for supplementing hard-to-find human talent — took on macroeconomic proportions. At the Agentforce 2.0 launch event, he argued that a global labor shortage is leading to reduced production and slowing growth. And he floated the idea that digital labor could represent as much as $7 trillion in potential economic upside.
The opportunity crystalized for Benioff after Salesforce decided to deploy AI agents on its help.salesforce.com website, which gets more than 60 million visits a year from customers seeking assistance. The site escalates the customer inquiry to a human rep if needed. However, since adding AI agents to the site in October, only half as many issues now require human intervention. Which means 83% of customer queries are now resolved without it.
In other words, digital labor now works seamlessly with human labor on help.salesforce.com to answer questions and fix things. “There are thousands of humans working with thousands of agents working with thousands of customers,” Benioff says.
That experience has contributed to this new way of thinking about the enormous potential of digital labor. “I’m not just managing human beings,” Benioff says. “I’m managing digital agents.” Which can only mean one thing: You need a platform to do it.
Integration with Slack, Tableau, MuleSoft
Salesforce describes Agentforce 2.0 as the first digital labor platform for enterprises, enabling businesses to build, deploy, and manage customized agents that perform complex, multi-step tasks. And it’s casting its net wider than sales, service, and marketing. It doesn’t want to be pigeon-holed as only a CRM company in the expanding agent market. “This is about the entire enterprise,” said Adam Evans, SVP and GM of Salesforce AI.
Part of what’s new with Agentforce 2.0 is that other elements of the Salesforce tech stack, including Slack, Tableau, and MuleSoft, have been tied in more tightly. The Slack integration in particular is sure to get attention given that millions of people (myself among them) use Slack every day. In a one-on-one demo, I saw six different agents embedded in Slack: a Benefits Specialist, IT Help Specialist, Product Specialist, Sales-Handoff Specialist, Sales-Deal Specialist, and Onboarding Specialist.
You can use predefined actions to collaborate and interact with Slack agents in myriad ways. Agents can create and update Slack Canvases; create and manage lists; build workflows; create new channels; search structured and unstructured data; summarize documents and conversations; and send direct messages on a user’s behalf.
Agents appear in Slack much like typical business users, so you can @ mention them in a message or channel just as you would a colleague, as illustrated in the screen shot below. In addition, a new enterprise search capability lets agents, with the appropriate permissions, draw from info available in Slack conversations.
Agents can comprise a combination of predefined and/or custom skills and actions. A new Mulesoft API Catalog and Topic Center make it easier to connect agents with APIs that inform their skills and actions. Also, a growing list of partners are adding their own skills to the Agentforce ecosystem, so there may eventually be hundreds or thousands of skills to mix and match for these new fangled digital workers.
The reasoning engine and data cloud
A few basic ingredients go into Agentforce agents: role, data, actions, channels, and trust & security. A key capability, and one that sets agents apart from an earlier generation of copilots, is their ability to reason. Salesforce has fine-tuned the Atlas Reasoning Engine that serves as the brains for its agents. So-called “advanced retrievers” now handle a wider variety of interactions and pull in data from more sources.
Here’s how the advanced retrievers worked in a demo with Claire Cheng, VP of AI engineering. Cheng posed a hypothetical question: “How will falling interest rates impact a financial portfolio?” The advanced retrievers explored deeper than a standard query, pulling relevant data and context-specific metadata from the Salesforce Data Cloud. The Atlas Reasoning Engine then assessed the response using an agentic loop for a more thoroughly-researched response.
Data professionals will also appreciate Agentforce 2.0’s enhanced RAG (retrieval augmented generation) with enriched indexing. To power the Atlas Reasoning Engine, Data Cloud can now augment RAG data with metadata from the Salesforce Platform. This lets agents apply business context, while boosting user trust through inline citations that point to the sources used.
Needless to say, trust in AI remains hugely important to Agentforce 2.0. Here’s how I described it in a Cloud Database Report post at the time of the original Agentforce announcement in September:
Salesforce has developed a “trust layer” to assuage such concerns. It bakes policy and principles into what AI agents are allowed to do and not do, a.k.a. guardrails. Capabilities include bias and toxicity detectors; security, privacy, and safety controls; audit trails; and the ability to “nudge” human intervention if needed. The trust layer serves as a gateway to LLMs to filter out unwanted behavior or inaccurate outputs. This helps reduce hallucinations.
Scaling to thousands of agents
Where is all of this leading? Potentially to thousands, millions, even billions of agents as the technology and best practices mature. It’s getting so easy that you can now create Agentforce agents using natural language descriptions, while pulling from a library of skills and actions. Agentforce 2.0 can even recommend skills that are needed by an agent.
So the barrier to entry is low and getting lower. All the more reason that I want to learn more about how agents will be managed at scale. I imagine big companies like FedEx (an Agentforce customer) might eventually deploy thousands of autonomous, digital agents. What does a CIO need to do that effectively?
Salesforce recognizes this requirement. For example, it recently introduced Agentforce Testing Center and enterprise-style monitoring, observability, and usage capabilities. There’s also an Agentforce ROI Calculator, so you can crunch the numbers for your business.
But there’s still a long way to go, as customers navigate and implement the new technology. “It’s the beginning of the beginning,” says Benioff.
Already, Salesforce is looking ahead to another round of Agentforce capabilities in the spring. The genie — a digital agent — is out of the bottle.