Salesforce unveils Agentforce to help create autonomous AI bots

Salesforce today released Agentforce, a new suite of low-code tools aimed at helping enterprises build autonomous AI agents for sales, service, marketing, and commerce use cases.

Agentforce, which has been in pilot phase for the past six months, combines three major Salesforce tools — Agent Builder, Model Builder, and Prompt Builder — to provide the necessary software development infrastructure to create these autonomous agents, according to the company.

Unlike chatbots, AI agents created via Agentforce will be capable of taking actions on their own, Salesforce claimed. The autonomous nature of such agents is a central facet of “agentic AI,” a rising enterprise strategy for transforming business processes by automating specific functions within those processes, without human intervention.

Salesforce’s journey to Agentforce

Salesforce previously enabled actions in conversational bots powered by large language models (LLMs) when it introduced Actions inside its Einstein Copilot in April this year.

Called “Copilot Actions” when released, these were a library of preprogrammed capabilities to help sellers benefit from conversational AI in Sales Cloud.

A top Salesforce executive explained to CIO.com that these “Actions” were basically workflows that could be built inside the Copilot via the Einstein 1 Studio set of low-code tools for creating, customizing, and embedding AI models in Salesforce workflows.

Business process automation (BPA) could be one action, for instance, the executive said; updating records, following up on emails, closing calls, forecasting guidance, and populating information are other examples.

These “Actions” could also tap Call Explorer’s retrieval augmented generation (RAG) functionality to help sales reps query prior call transcripts captured in Einstein Conversation Insights, including asking questions about customer sentiment.

The CRM provider first announced Einstein Copilot a year ago, with Einstein 1 Studio coming in March of this year as part of its bid to become the platform of choice for building AI assistants for business, especially for sales, service, and marketing use cases.

Agentforce takes that strategy — and ambition — a step further.

What makes Agentforce agents stand out?

AI agents created via Agentforce differ from previous Salesforce-based agents in their use of Atlas, a reasoning engine designed to help these bots think like a human being, according to Salesforce.

“We’re going from content generation and analysis to automating actions,” Clara Shih, CEO of Salesforce AI, said in a statement.

“In the generative phase, you might ask a copilot to write an email for you to a customer. In the agentic phase, you can ask a harder question: ‘What should I do with all of my customers?’ Maybe it’s email, maybe it’s picking up the phone and calling, maybe it’s sending a text message,” Shih said. “That’s really what agents can do: They can take a higher-order question, break it down into a series of steps, and then execute each of those steps.”

Explaining further, the company said its Atlas reasoning engine starts by evaluating user queries and refining them for clarity and relevance, after which it retrieves the most relevant data and builds a plan for execution.

The process then refines the plan further, ensuring it’s accurate, relevant, and grounded in trusted data, according to Salesforce. Based on this process, Agentforce agents are able to reason, make decisions, and complete business tasks autonomously, while delivering factually accurate results.

Salesforce also said it would rebrand Einstein Copilot as an Agentforce-developed agent, as Copilot has been upgraded to now be capable of retrieving data, reasoning, building a plan, and taking action.

Out-of-the-box autonomous agents

In addition to enabling enterprises to develop their own agents through Agentforce, Salesforce is also releasing out-of-the-box agents, including a service agent and agents for buyers, shoppers, merchants, and campaigning.

The service agent replaces traditional chatbots with AI that can handle a wide range of service issues without preprogrammed scenarios, improving customer service efficiency, according to the company.

Merchant and buyer agents aim to help organizations enhance customer experiences, Salesforce said. To do so, the merchant agent assists merchandisers with site setup, goal setting, personalized promotions, product descriptions, and data-driven insights; the buyer agents helps customers find products, make purchases, and track orders via chat or within sales portals.

Another agent, dubbed personal shopper, acts as a digital concierge on ecommerce sites or messaging apps, offering personalized product recommendations and assisting with search queries.

Campaign optimizer automates the full marketing campaign lifecycle, using AI to analyze, generate, personalize, and optimize campaigns based on business goals, the company said.  

Last month, the company released two autonomous AI agents — Einstein Sales Development Rep (SDR) Agent and Einstein Sales Coach Agent — built on the Agentforce platform.

Workflows, availability, pricing

What seems to have not changed is the ability of these new autonomous agents to be weaved inside workflows just like Einstein Copilot.

“With Salesforce Flow, MuleSoft, and Apex methods, customers can easily extend the functionality of Agentforce by tapping into workflows and actions that are already built and optimized,” the company said in a statement.

“They [customers] can also use these familiar building blocks to create new automations for Agentforce as well,” enabling enterprises to capitalize on their existing investments in automation while scaling new capabilities, Salesforce explained.

Agentforce for Service and Sales will be generally available Oct. 25, 2024. The company did not clarify when the suite of tools will be available for marketing and commerce use cases.

Furthermore, Salesforce said some components of its Atlas Reasoning Engine will be launched in February 2025.

The cost of the using the service starts at $2 per conversation, though the company said standard volume discounts may apply.

Last month, the company had said that it was mulling a new consumption-based pricing model for its AI agents.

Ritu Jyoti, general manager of research at IDC, provided comment on Salesforce’s shift on AI agent pricing.

“Since it is a start for their offering where the usage will be fragmented and will take some time to scale adoption, I think consumption-based usage is fine with volume discounts,” she said.

“Given its impact on improving the deflection rates, which is an important priority for majority of customers, $2 per conversation may not be too high. I would expect it to morph into subscription plus usage in the long term, where output accuracy coupled with the outcome targeted will be key in driving value,” Jyoti added.

Along with Agentforce, Salesforce has also launched an Agentforce Partner Network, which includes the likes of AWS, Box, Google, IBM, and Workday.

These partners have already built 20 agents and agent actions that will be made available through the Salesforce AppExchange for enterprises to use, the company said.