Salesforce launched Agentforce 2.0 in a bid to move its agentic AI platform to more teams, workflows and markets. Agentforce 2.0 includes Slack integration, more workflows across the enterprise and is generally available in February.

The rollout comes just weeks after Agentforce made its debut at Dreamforce.

Agentforce 2.0's debut also includes a new tagline dubbing the effort a "digital labor platform" that can move beyond Salesforce's core sales, service and CRM borders. To expand Agentforce's reach, Salesforce is relying heavily on MuleSoft, the company's API platform. With MuleSoft, Agentforce will be able to enable actions across 40 enterprise systems via MuleSoft for Flow, surface APIs as actions and add metadata to APIs.MuleSoft for Flow will be in pilot Feb. 25 and surfacing APIs as actions and metadata additions to APIs in Topic Center will be available on the same date. MuleSoft has become critical to Salesforce's plan to expand into new agentic AI use cases.

Salesforce is also adding new pre-built skills that will be released monthly. New pre-built skills in Agentforce will include sales development and coaching, field asset management, commerce merchant, Slack engagement, predictive insights and industry-specific skills. Agentforce 2.0 will also include an expanded roster of third-party skills. Salesforce's Agentforce 2.0 is a fast follow to its Dreamforce launch of its agentic AI platform. 

Salesforce CEO Marc Benioff said:

"As a CEO, I'm not just managing human beings, but I'm also managing agents. There is an agentic layer around the support today for Salesforce. It's not some vision fantasy in the future idea, it's what is happening right now."

Benioff said agentic AI is enabling digital labor, robots and voice as a user interface. "We've already crossed this bridge into the new world of digital labor," he said. 

Here's how Salesforce sees Agentforce 2.0 expanding into new areas and providing digital labor.

Salesforce's approach with Agentforce 2.0 rhymes with ServiceNow's focus on AI-enabled workflow orchestration. Where ServiceNow started in IT and expanded to other areas including ERP, HR, CRM and any other function, Salesforce is starting with CRM and expanding out into HR, finance and engineering.

Slack search

Salesforce also appears to be leveraging Slack as the collaboration glue for Agentforce 2.0. On Jan. 25, Agentforce will be available in Slack, integrate into Slack enterprise search and feature prebuilt actions. In Slack, Agentforce can take action including creating a Canvas, updates, adding channels and sending direct messages.

Key points about how Slack search is used in Agentforce:

  • Slack's search system ensures that a message is visible only to the right users by indexing messages based on channel types and applying visibility filters at the time of query. For instance, public channels are visible to all, but private channels only apply to members of that channel. Search queries are indexed to workspace ID in the public channel scenario, but the index avoids listing every member ID in the private channel scenario. 
  • File visibility is also determined by channels and members. Files in public channels will be indexed by workspace ID, but private channels are restricted to channel membership. 
  • Slack's search functions are enhanced by retrieval-augmented generation (RAG) techniques. 
  • Slack search actions are also controlled by visibility access, admin controls and minimized data sharing. 
  • Slack search also has FedRAMP certifications, enterprise key management and custom retention policies. 

Atlas Reasoning Engine details

Salesforce also enhanced its Atlas Reasoning Engine with features that will be generally available Feb. 25. These enhancements include:

  • Advanced Retriever, which can be used to improve answers.
  • Enriched Indexes that will use advanced RAG and added metadata.
  • And Inline Citation, which will improve accuracy and transparency to responses.

The company also added some details to how Atlas works. According to Phil Mui, Ph.D., SVP of Technology, Head of Products and Engineering, Salesforce AI Research and Agentforce, Atlas combines models, data, business logic, events and workflows into one compound system. Atlas aims to understand the nuances of user queries and contexts. 

Mui argued that DIY assistants that are large language model based are rapid, but are focused on basic tasks (System 1). Atlas is an evolution because it can understand and think through a situation (System 2). 

Atlas retrieves the most relevant data based on processes and then refines the query with RAG and additional context and metadata. This step-by-step approach to Atlas' analysis grounds answers and decisions using Data Cloud. 

Salesforce noted that Atlas' move to System 2 thinking is a win, but future progress will incorporate agentic AI and coordinating specialized agents.  

A faster cadence, bigger market

Salesforce President and Chief Operating Officer said at a recent Barclays investor conference that the company is stepping up its innovation cadence with more frequent releases. Benioff also noted agentic AI is expanding Salesforce's market. 

In February, Agentforce 2.0 will feature a bevy of new skills beyond those just outlined including citation support, supervisory LLMs, token streaming, scheduling skills and multilingual support.

"We've prided ourselves on three upgrades a year for 25 years. This (Agentforce) is measured in weeks and so the velocity is really incredible," said Millham. "We feel that urgency by the way. We really feel like we need to be moving quickly at this opportunity."

And the opportunity is large as Google Cloud, AWS and Microsoft Azure all have agentic AI efforts underway. 

Millham said Salesforce with Agentforce can expand the company's total addressable market dramatically. He explained:

"We're going to provide digital labor to our customers now. And when you start to look at this through the lens of just software, the size of that measured in billions. When you start to think about what the opportunity is for labor offset, you measure in TAM and in the form of trillions. That is the opportunity that we're contemplating out there.

We talk to CEOs and CFOs, they say, this is great. I can take some cost out of my business in the call center, for example, which is fine.

We also can have it from a growth perspective. Hey, over the next five years, I'm growing at 8% on top line and my cost of people is flat. That's a pretty good proposition and people can do the math very quickly on why they should be investing in Agentforce."

Millham said customers are asking about applying Agentforce to other areas such as HR. At the conference, Millham was asked about how Agentforce will be priced. He said pricing for Agentforce will start to resemble platform plays where you have prices for the platform and then universal credits that will be consumed across workflows. In other words, Salesforce's pricing model will likely rhyme with ServiceNow's approach going forward.

Constellation Research's take

Constellation Research analyst Holger Mueller said:

“Salesforce is demonstrating unseen product velocity with the launch of Agentforce 2.0. Mark Benioff is the Jensen Huang of SaaS by accelerating product innovation cycles. Bringing Agentforce to where work happens -- to Slack is critical. Even more will be the necessary integration into Teams / Outlook (no date so far). Equally critical for CxOs will be - which other SaaS vendors will play along with Salesforce - as Salesforce opens Agentforce to ISVs. Will integration happen on a data level, an API level or even an agent level - will be a key AI success question for 2025 - not just for the industry - but for enterprises.”

Constellation Research analyst Martin Schneider said:

"One of the most interesting facets of the Agentforce 2.0 launch is the apparent rebranding of agentic AI into 'digital labor.' This development seems interesting when you think about the existing ambivalence towards AI from a lot of workers and the fear that it will displace labor rather than augment human capabilities. From the business side, the digital labor platform messaging of course implies cost savings and efficiency, but I think Salesforce needs to be careful in how they put this message across certain audiences.

From a technical side, Salesforce seems to be doing the right things in terms of enabling nontechnical users to get started with agentic AI, not just from putting the work in around Data Cloud to power more effective insights, but also pre-integrating with MuleSoft to power the task management and workflow associated with AI agents taking autonomous action. In addition, the new skills concept gives both technical and non-technical users a lot of components that are preconfigured towards building more effective agents out-of-the-box.

Out-of-the-box agents now run across all Salesforce clouds, giving users some fast and useful proof of concepts as they dip their toe into the water of agentic AI. I expect Salesforce's partner ecosystem to go even further and offer even more AI agents that cover even more complex use cases that touch even more systems than CRM and ERP. 

But in the end, the two important factors will be both the quantity and quality of usable data that Salesforce users have on hand to power AI agents. Historically, CRM data can be both incomplete and inaccurate. As AI use cases call for more and more queries into the core Salesforce platform, performance and availability may become an issue as some users are already citing that the more they leverage complex AI use cases, the number of round trips between the Atlas reasoning engine, other data systems, and Salesforce and data cloud are causing some performance issues.

The availability of Agentforce in Slack will be strong bridge between the agentic-to-human handoff, allowing for more seamless workflows, and a better ability for humans to monitor AI agents inside a tool like Slack that is m ore conducive to these actions than a traditional CRM UI."

Business Research Themes

Data to Decisions, Future of Work, Innovation and Product-led Growth, Marketing Transformation, Matrix Commerce, Next-Generation Customer Experience

Roles

Chief Information Officer

Tags

AI agents, #Salesforce, #AI #artificialintelligence #machinelearning #ML #generativeAI