A First Look at Salesforce’s New Agentforce Platform Before Dreamforce

Agentforce is Salesforce’s latest “hot” innovation, teased already during their Q2 ’25 earnings call at the end of last month. Now, there’s been more light shed on why Agentforce is such a big deal. In yesterday’s briefing call, the energy radiating from Marc Benioff (Salesforce’s CEO and Founder) was palpable.

In basic terms, Agentforce allows organizations to build and manage autonomous agents for tasks across various business departments – not exclusively for customer service use cases. “Agentforce in every app” is the first of the announcements, with the Atlas reasoning engine, Agent Builder, and the Agentforce partner network also being unveiled. 

We’ll cover the latest announcements in the run-up to Dreamforce, and explore the potential impact it could have (spoiler: there’s more to this than meets the eye).

What Is an Agent? 

A good place to start is to outline the five attributes of an Agent. Building one with Agentforce will need to have the following defined: 

1. Role: What are the jobs and tasks that need doing? 
2. Data: What can they access? This applies to both structured data, and unstructured data (e.g. call transcripts, Slack conversations, emails, etc.)
3. Actions: What capabilities do they have? This is not limited to just answering questions, but being able to perform based on business logic – that’s the flows, Apex code, MuleSoft API calls, etc. that exist in your org and have been battle-tested through time with employees. 
4. Guardrails: What shouldn’t agents do? This leverages the Einstein Trust Layer, which delivers data privacy (e.g. data masking, zero retention prompts, prompt injection defense, and ethical guardrails (e.g. toxicity filters) and the ability to monitor the AI. Guardrails go further – in fact, you could consider anything in your Salesforce org to be a guardrail (role permissions, data model, business logic). 
5. Channel: Where do they work? This details the mediums through which the agent can communicate with customers and employees, such as SMS, WhatsApp, Slack, or voice.  

The key takeaway from this section is that most of these items and infrastructure already exist within your Salesforce org. Through Agent Builder, you can tap into these building blocks. 

Agents vs. Chatbots

Another useful piece of context is to answer “How are agents different from chatbots?”

Traditional bots are like pre-programmed assistants. They can only answer questions that they’ve been specifically told about – if they don’t know, they will give you a generic answer. 

Unlike conventional systems that handle only pre-programmed queries, agents use large language models (LLMs) and generative AI to understand the full context of customer messages. It autonomously determines and executes appropriate actions, such as processing requests or resolving issues, based on your company’s CRM data and brand guidelines. 

Agentforce “In Every App” 

Agentforce is available in every app (i.e. Salesforce “louds” in their product suite) – so, that’s Service Agents, Sales Agents, Marketing Agents, Commerce Agents, and Platform Agents. 

Atlas Reasoning Engine 

As the “brain” behind Agentforce, Atlas generates a plan based on what the role is trying to do, then evaluates and refines its plan (if required, it can loop to pull in additional data). When taking an action, it will look at what business process (e.g. flows) should be used. Finally, it engages with the customer or employee based on their channel preference.   

The diagram below is helpful in understanding the core components of the process – the point being to, ultimately, generate precise and factual results. 

The Atlas Reasoning Engine is worth looking into further, being an agentic system (versus an assistive system), which means it has more agency in actions that it performs on behalf of the user.

Agent Builder

The newest in Salesforce’s “family” of low-code builders, Agent Builder joins Prompt Builder and Model Builder to form the Agentforce Studio (formerly Copilot Studio). 

• Prompt Builder to create prompt templates. Choose how they want to “ground” the prompts in Salesforce data, and activate for users quickly and easily. 
• Model Builder to train, or bring in your own AI models.
• Agent Builder to create and customize agents, and enable agents with automation, APIs, and code on the Salesforce platform. 

Agentforce Partner Network

The final of the four announcements is the partner network of leading software vendors that extend agent-building capabilities, such as adhering to the same ethical guardrails, data sharing, and actions.

Example Agent: Sales Agent

Sales Agent, which comes in two forms (Sales Development, and Sales Coach) enhances sales teams by automating and personalizing tasks such as nurturing leads, coaching reps, and managing inbound inquiries. Configured easily through a user-friendly interface, it allows you to set permissions, customize interactions, and focus on specific leads based on their value. 

It uses uploaded sales materials and CRM data to send personalized introductory emails, handle objections, and suggest meeting times.

The governance dashboard helps monitor performance, ensuring that sales goals are met and providing insights into each lead’s status.

What’s the Big Deal? 

We’ve already compared how agents go far beyond traditional chatbots, offering flexibility with the ability to set guardrails and avoid complexity and time when building. 

As part of Agentforce, Salesforce has rewritten Customer 360 into a new platform called Foundation (available next week). Beyond being a new user interface, it’s more powerful in terms of processing, and every acquisition Salesforce has made will now be amalgamated into the platform. This delivers better access to data, and in turn, more accurate AI. Initial figures that customers have reported are around 90% accuracy and relevancy, and promoter scores that have doubled, or even tripled. 

Plus, it’s always worth considering Salesforce’s offerings compared to a DIY (do it yourself) approach. Salesforce believes in democratizing AI with these built-in innovations, which are low-code for the most part (therefore not requiring highly skilled people in-house) and deliver faster time to value. 

We’ve probably all heard some version of the cautionary tale whereby some organizations spend a huge sum of money, years implementing AI, and not even seeing much impact from it. One goal at Dreamforce will be to prove that these cutting-edge capabilities can be deployed in a rapid timeframe. 

Availability and Pricing

Agentforce for Service and Sales will be generally available on October 25, 2024. Some components of the Atlas Reasoning Engine launch in February 2025. Agentforce pricing starts at $2 per conversation, with volume discounts on offer.

At $2 per conversation, Salesforce predicts that customers will see huge ROI. Agentforce agents are more cost-effective than back-and-forth conversation with a human agent in freeing them up to deal with more nuanced queries. 

Summary

Salesforce’s Agentforce emerges as a game-changing innovation that seamlessly blends AI with human capabilities to transform various business aspects.

As Dreamforce unfolds, more exciting announcements are anticipated, potentially unveiling the far-reaching impact of Agentforce.