In the landscape of customer relationship management, Salesforce has introduced AgentForce. A groundbreaking AI powered framework designed to revolutionize automation and decision making within the Salesforce ecosystem.

In CRM, Salesforce has launched AgentForce. It's a new AI tool that aims to transform automation and decision-making in its ecosystem.

What is AgentForce?

AgentForce is Salesforce's AI framework. It helps users create and deploy AI agents in the Salesforce ecosystem. It provides a structured approach to use AI in Salesforce orgs. It covers automation and decision-making.

Core Components of AgentForce

1. Actions

The foundation of AgentForce lies in its actions. They serve as the basic building blocks of the framework. These actions:

• Represent specific tasks or operations.
• Can be either custom-built or pre built Salesforce actions.
• Form the foundation for more complex operations

2. Flow Actions

Flow actions are crucial for orchestrating the system's operations:

• Manage the sequence of actions
• Determine how actions connect and interact
• Handle data flow between different components
• Control execution paths based on conditions and results

3. Apex Actions

The programming backbone of AgentForce includes:

• A comprehensive programming framework
• Custom logic implementation
• Complex business rules handling
• Interfaces with Salesforce data and ability to interface with external systems

4. Prompt Actions/Builder

This component sets the stage for agent operations by:

• Defining initial input or triggers
• Setting context for agent operations
• Including specific instructions or parameters
• Determining agent interpretation and response protocols

Agent Actions

These specialized components are designed to:

• Execute specific AI powered tasks
• Uses Flow/Apex/Prompt actions.
• Process input data and generate results
• Interface with other Salesforce components
• Chain together for complex operations

Topics

Topics provide context and categorization for agent operations:

• Help organize different types of tasks
• Enable better routing of requests
• Facilitate proper agent action selection
• Improve response accuracy

Topic Classification

Classification within Topics:

• Categorizes incoming requests
• Determines appropriate response paths
• Enables proper handling of different scenarios

Topic Instruction

Instructions within Topics:

• Provide specific guidance for handling requests
• Define response parameters
• Set boundaries for agent operations
• Grounding the prompt

Overall Agent Flow

• A prompt initiates the process through agent.
• Topics classifications decided which topic is best for the request
• Instructions in topic guide agent action selection
• Selected agent action executes actions
• Actions are coordinated through flows/apex class or prompts
• Actions response returned to agent.

Available Agent Types

• Salesforce Copilot Agent

The Copilot agent serves as your internal AI assistant within the Salesforce ecosystem. It's easily accessible through the top right menu on any org page. It's designed to do many organizational tasks. It can be customized through training to meet specific business needs.

• Digital Agents

These versatile agents are built for external channel integration. They can offer comprehensive task management capabilities across various platforms.

Current Challenges and Limitations

While AI agents bring impressive capabilities, we've identified several key challenges:

Custom Field Recognition Issues

• The systems currently struggle with identifying specific custom fields
• Notable confusion between Ship Date and Process Start Date
• Difficulty distinguishing between Estimated and Actual Return Dates for loans
• Solution needed: Development of custom action handlers

Generic Term Mapping Challenges

• Gap between everyday language and specific business categories
• System doesn't naturally connect terms like "Loans" to "Equipment Loans"
• Similar disconnect between "Orders" and "Equipment Orders"
• Required: Custom actions for improved categorization

Limited CRUD Operations

• Current limitation: Agents lack out of the box update and delete capabilities
• Need for expanded functionality to enhance versatility

Successfully Implemented Features

Habilelabs has successfully implemented several key functionalities:

• Natural language querying for record access
• Comprehensive record summarization
• Case and user owned record summary generation
• Object/order updates via agents
• New object creation capabilities
• Reminder email system
• Email follow up
• User engagement summary

Next set of tasks

Enhanced Communication Integration

• Exploring mobile app integration
• WhatsApp connectivity possibilities
• Other communication channel expansions

Security and Permissions

• Deep dive into object permission frameworks
• Enhanced security setup implementations

Knowledge Base Enhancement

• Integration of external knowledge bases
• Leveraging Salesforce DataCloud capabilities