Enterprises witnessed a plethora of AI advancements in the last couple of years and 2024 stands out in terms of AI developments in the CRM space. The reason: AgentForce by Salesforce.
What is Agentforce?
Agentforce is a trusted, natural language conversational AI Agent that helps business users and sales teams, service agents resolve customer issues, answer queries and take autonomous actions.
Behind the magic
The core components working behind Agentforce are Topics, Actions, The Powerful Prompt Builder, and the core Salesforce platform features — Lightning Flows and Apex.
Actions and Topics:
Actions are what an agent can do when appropriate—such as summarizing information, getting answers from the Knowledge Base, checking the status of an order, or drafting emails. Some common Salesforce actions—such as finding a record by name, drafting or revising a sales email, and summarizing a record—are available out of the box as standard actions that one can use immediately.
However, the true power of Agentforce lies in its ability to address specific business needs through custom actions built with tools such as Prompt Builder, Apex, and Flows. These custom actions can range from initiating an order return to logging IT tickets, checking inventory levels, or creating and modifying invoices.
Topics organize Agentforce’s actions. They help agents make more accurate decisions and generate more relevant, predictable responses. A Topic includes related actions and instructions, guiding agents on what to do, what is allowed, and what to avoid. This system nearly eliminates irrelevant or generic answers, acting as a form of “Guardrail.”
For example, a Topic called Case Management might include actions that help service representatives get up to speed on their day, find open cases to work on, determine critical priorities, create to-do items, contact clients for more information, and log calls.
The Atlas Reasoning Engine:
The ARE (Atlas Reasoning Engine) acts as the brain of Agentforce. It interprets the user’s request, classifies it into a topic and builds a plan to accomplish the goal of the user by launching the right topics and actions to achieve the goal.
How is Agentforce different from Einstein Copilot?
At first glance, Agentforce may seem similar, but it is far more advanced than Copilot. Agentforce operates more independently handling tasks and taking decisions autonomously and can process natural language in a better way without the need for constant human intervention. Agentforce, powered by ARE, processes both structured and unstructured data—such as PDFs and call transcripts. While Copilot had limited configurability, Agentforce allows extensive customization, enabling enterprises to tailor AI to their diverse use cases across areas like Sales, Service, Media, Commerce, Education, Finance, and Health, making it truly “What AI was meant to be.”
While Copilot was assistive, Agentforce is truly agentic. With its RAG (Retrieval-Augmented Generation) capabilities, it continually stays updated with new data from its sources.
Trust and AI:
Feeding enterprise data into AI can deliver up to a tenfold acceleration in processes and substantially increase profits. However, data security and trust remain crucial considerations.
“What safeguards does Agentforce provide to ensure that data remains secure as organizations leverage its capabilities for success?”
Like other Salesforce’s features, Agentforce is integrated with the Einstein Trust Layer, the secure AI Architecture natively built into Salesforce. The Trust layer lets enterprises benefit from generative AI without compromising their customer data.
The Trust Layer ensures:
- Data grounding: The Trust Layer ensures that generative prompts are grounded and enriched in trusted company data.
- Zero-data retention: The company’s data is never retained by a third-party LLM provider.
- Toxicity detection: Potentially harmful LLM responses are detected and flagged.
- AI monitoring: AI interactions are captured in event logs, giving you visibility into the results of each user interaction.
All data accessed by agents, including personally identifiable information (PII), is protected in transit and isn’t stored or used for training purposes by external LLM providers, as part of Salesforce’s strict zero-data retention policy.
The Agentforce platform includes ethical guardrails to minimize AI hallucinations and security guardrails to protect against threats and malicious attacks, such as prompt injections. Through topic instructions, users can set boundaries, establish contexts, and define an agent’s behavior
Pre-requisites to get started with Agentforce in your Org:
Data Readiness with Data Quality and Accessibility: Cleaning, organizing, and maintaining accurate customer records without which using AI will not give the expected results or actionable insights.
Network Readiness: Strengthen the organization’s network infrastructure with sufficient bandwidth, low latency, and secure configurations.
Security and Compliance Readiness: This is important especially in the regulated sectors like healthcare or finance. Organizations need to be clear about the regulatory standards and define them in the org by addressing compliance at the onset. This builds a strong foundation for a trustworthy, compliant Agentforce.
What is new and upcoming in Agentforce?
Agentforce has already proven its value for many enterprises, providing exceptional capabilities and driving significant productivity gains. Now, with the release of Agentforce 2.0, Salesforce is introducing the following features:
Agentforce 2.0:
- New Skills for Sales Teams like Development and Coaches, Merchant Skills for Commerce, Scheduling skills for service engagements and new skills for field service workers.
- Take Action Across Any App or Workflow with MuleSoft: MuleSoft empowers Agentforce to streamline business workflows by simplifying low-code workflow creation across any system. With MuleSoft for Flow, teams can quickly build multi-system workflows using pre-built connectors. The MuleSoft API Catalog enables builders and Salesforce Admins to view, discover, and manage APIs across Salesforce, MuleSoft, Heroku, and external services from one location. Meanwhile, the MuleSoft Topic Center infuses Agentforce metadata into every API, automatically converting it into an Agentforce skill or action, making each API “agent-first” by default.
- Tableau Skills for Analytics and Insights: New Tableau Topics and Actions deliver data visualizations and predictions for deeper understanding of agent responses and accurate, business context-rich answers using Tableau Semantics. This unlocks new conversational analytics use cases, further lowering the barrier to data access for everyone.
- Agentforce Recommended Skills: Enables enterprises to build agents in seconds using natural language. The Agent Builder automatically composes new agents for work by generating relevant topics and instructions, while also integrating any pre-existing actions from the library.
- Agentforce in Slack: Agentforce 2.0 lets teams bring the agentic AI into any Slack conversation
Key Dates:
- Agentforce 2.0’s full release will be generally available in February 2025
- Specific features for 2.0 are being released in advance – beginning today
- Skills for Tableau will be generally available from December 18, 2024
- Agentforce in Slack and other Slack functions will be generally available in January 2025
- Natural language creation of agents in Agent Builder will be available in January 2025
- MuleSoft for Flow, MuleSoft API Catalog, and Topic Center will be available in February 2025
- Enhanced reasoning and RAG will be generally available in February 2025
In summary
Enterprises already on the Salesforce platform—or just adopting it—can now leverage these cutting-edge AI capabilities to gain immediate value in their Salesforce org. By democratizing AI through low-code innovations, Salesforce ensures faster time-to-value. ABSYZ, a leading Salesforce Platinum Partner, fully embraces these advancements—delivering tailored AI solutions that accelerate innovation and drive measurable business outcomes.
Authors: Subrahmanyam Daita
