Issacc.com

Agentforce Specialist Exam Prep - Agentforce Concepts

11 min read

Agentforce Concepts consists of 30% of total score in Salesforce Agentforce Specialist Exam. The topic covers agents & reasoning engines, actions & topics, managing & monitoring agent adoption, user security, testing center, and deploying Agentforce to Production.

NOTE

Most of the content in this work was generated with the assistance of AI and carefully reviewed, edited, and curated by the author. If you have found any issues with the content on this page, please do not hesitate to contact me at support@issacc.com.

🤖 Agentforce Concepts — Comprehensive Summary

Goal: Understand how Salesforce Agentforce works — including its reasoning engine, agent actions, topics, analytics, security, testing, and deployment.

🧠 Agents & Reasoning Engine

🔍 Overview

Agents like Agentforce (Default) communicate with users and take actions through the Reasoning Engine and Large Language Model (LLM) — mainly OpenAI GPT-40.

⚙️ How the Reasoning Engine Works

  • Orchestrates topics and actions during a conversation.
  • Interprets user input, identifies intent, and builds plans to meet goals.
  • Works closely with the LLM to execute actions and generate responses.

🔄 Interaction Flow

  1. Identify user intent from natural input.
  2. Generate dynamic plan containing relevant agent actions.
  3. Execute actions and perform necessary LLM calls.
  4. Generate responses (LLM + system messages).
  5. Continue the conversation seamlessly.

📋 Reasoning Engine Core Functions

  • Interprets triggers or user requests.
  • Classifies utterances into topics.
  • Builds and executes plans iteratively.
  • Launches appropriate topics & actions.
  • Produces contextual responses.

🧾 Event Logs

Use the Events panel in Agent Builder to:

  • Debug agent behavior.
  • View session logs.
  • Analyze execution of actions.

⚠️ Considerations

AspectDetails
Planner Service CallsUses OpenAI GPT-40
Custom ActionsCan use any Salesforce-managed LLM
LogsViewable in Agent Builder → Events Panel

🧩 Agent Actions & Topics

🎯 Agent Actions

Actions are individual tasks that an agent performs on behalf of users.

Types

TypeDescription
Standard ActionsPredefined tasks included by default (e.g., Summarize Record, Draft Email).
Custom ActionsExtend capabilities to handle specific business needs via Apex, Flow, or Prompt Templates.

🧱 Reference Action Types

  • Apex Class (invocable)
  • Autolaunched Flow
  • Prompt Template

🧭 Action Assignment

  • Actions must be assigned to topics for the agent to use them.
  • One action can belong to multiple topics.
  • Managed in Agent Builder → Topics Panel (requires deactivation to modify).

🗒️ Action Instructions

When creating custom actions:

  • Define what the action does and when/how to use it.
  • Optionally require user input or mark parameters as visible in conversation.
  • Use examples and utterances for better LLM understanding.
✅ Best Practices

Include:

  • 1–3 sample utterances.
  • Dependent actions if any.
  • Input/output examples for context.

🗂️ Agent Topics

ConceptDescription
DefinitionA topic is a category of actions related to a job the agent performs.
ComponentsActions, Classification Description, Scope, and Instructions.
PurposeOrganizes agent behavior, sets context, and defines conversation boundaries.
CustomizationStandard topics can be used, edited, or created from scratch.

🧩 Managing Topics

  • Add/remove topics via Agent Builder.
  • Standard topics are in the Asset Library.

🧰 Examples of Standard Actions

ActionDescriptionExample Utterance
Answer Questions with KnowledgeResponds using Salesforce Knowledge Articles.“What’s our return policy?”
Create Close PlanGenerates sales plan to close opportunities.“Write a close plan for Edge Solar Panel.”
Draft or Revise EmailComposes or updates email drafts.“Write a follow-up email to Sean Forbes.”
Find Similar OpportunitiesReturns deals similar to a given opportunity.“What other deals like Edge Installation have we won?”
Query RecordsRetrieves records based on criteria.“Show me open cases for Edge Communications.”
Summarize RecordSummarizes record data via prompt templates.“Summarize the University of Arizona account.”

📊 Managing & Monitoring Agent Adoption

📈 Agentforce Analytics

Provides dashboards & reports on:

  • Agent adoption & usage.
  • User feedback.
  • Performance metrics.

💬 Utterance Analysis

  • View what users ask the agent.
  • Analyze which requests the agent can handle.
  • Identify training opportunities for improvement.

🧾 Event Logs

  • Access via Agent Builder → Event Logs tab.
  • View all user interactions for debugging and optimization.
  • Enable “Enrich Event Logs with Conversation Data” to store detailed session data (retained 7 days).

🧱 Data Cloud Integration

  • Uses Copilot for Salesforce Apps Events Data Model Object to store event data.
  • Dashboards built using Data Cloud objects.

🔐 Agentforce User Security

🧭 Overview

Agentforce security is managed via permission sets that define user and agent access.

🧑‍💼 Permissions for Creating & Using Agents

TaskRequired Permissions
Create Agentforce (Default)Manage AI Agents + Manage Agentforce Default Agent
Create Other Agents (e.g., Sales Coach)Manage AI Agents + specific permission (e.g., Manage Sales Coach Agent)
Use Agentforce (Default)Access Agentforce Default Agent permission set group

👤 Agent User Record

  • Each agent must have a user record assigned to a permission set with the Agent User license.
  • Example: Agentforce Service Agent User permission set.

🧾 Example Permission Sets

Permission SetAgent TypeDescription
Use Sales Coach AgentSales CoachAllows users to access Sales Coach.
Manage Sales Coach AgentSales CoachAdmin configures Sales Coach.
Agentforce Sales CoachSales CoachAssigned to Sales Coach user record.
Use Agentforce SDR AgentSDRGrants user access to SDR agent.
Configure Agentforce SDR AgentSDRAllows admin to manage SDR.
Manage Agentforce Service AgentService AgentAllows admin to build/manage Service Agents.
Agentforce Service Agent UserService AgentAssigned to Service Agent user record.
Service Planner BuilderService PlannerAllows admin to manage Service Planner.

⚠️ Security Best Practices

TIP

Principle of Least Privilege: Only grant necessary access.
Two-Factor Authentication: Required for sensitive actions.
Scope Limitation: Private actions should confirm user identity before execution.

🧪 Agentforce Testing Center

🧰 Purpose

Allows batch testing of a large number of utterances to ensure accuracy before activation.

🧾 Process

  1. Create CSV file with test cases (use template).
  2. Create test for the agent in sandbox.
  3. Upload CSV and run test.
  4. Analyze results — view passed/failed utterances.

⚙️ Capabilities

  • Test up to 1,000 utterances per test.
  • Run up to 10 tests in a 10-hour window.
  • Supports multiple topics & actions simultaneously.
  • Failed utterances can be re-tested in Conversation Preview.

⚠️ Considerations

AspectDetails
ConsumptionUses Einstein Requests + Data Cloud credits
EnvironmentOnly in sandbox (prevents CRM data change)
Add-ons RequiredEinstein for Sales / Service / Platform

🚀 Deploying Agentforce to Production

🧭 Overview

After successful testing, deploy the agent from sandbox → production using:

  • Change Sets or
  • Metadata API / Salesforce CLI

🔄 Change Set Deployment Steps

  1. Create Outbound Change Set in sandbox.
  2. Add all relevant metadata components.
  3. Upload to production as Inbound Change Set.
  4. Validate and deploy.

🧩 Required Metadata Components

ComponentType / Example
Agent PlannerGenAiPlanner
TopicsGenAiPlugin
Agent ActionsGenAiFunction
Prompt TemplatesGenAiPromptTemplate
Flows & Apex ClassesReferenced logic components
Einstein Bot & VersionRequired for conversational agents
Embedded Service DeploymentFor Service Agents (Experience Cloud)

🔑 Activation & Communication

  • After deployment, activate the agent in Agent Builder.
  • Notify stakeholders and provide training/documentation.

✅ Key Takeaways

ConceptSummary
Reasoning EngineCore planner service orchestrating LLM + actions.
Agent Actions & TopicsModular, reusable tasks organized under topics.
AnalyticsDashboards for adoption, feedback, and performance.
SecurityRole-based permissions + least privilege principle.
Testing CenterBatch test up to 1,000 utterances for efficiency.
DeploymentUse Change Sets / Metadata API; activate post-deployment.

📈 Flow Charts

1) Agent ↔ Reasoning Engine: End-to-End Flow

flowchart TD
    A["User request"]
    B["Reasoning engine planner"]
    C{"Classify into topic?"}
    D["Select topic"]
    E["Ask for clarification"]
    F["Select actions"]
    G{"Inputs complete?"}
    H["Collect data from user"]
    I["Execute actions"]
    J["LLM call"]
    K["Generate response"]
    L{"Goal met?"}
    M["End or continue"]

    A --> B
    B --> C
    C -->|Yes| D
    C -->|No| E
    D --> F
    F --> G
    G -->|No| H
    G -->|Yes| I
    I --> J
    J --> K
    K --> L
    L -->|No| B
    L -->|Yes| M

2) User Request Processing

flowchart LR
    A["Identify intent"]
    B["Generate plan"]
    C["Execute actions"]
    D["LLM call"]
    E["Generate response"]
    F["Continue conversation"]

    A --> B
    B --> C
    C --> D
    D --> E
    E --> F

3) Topics and Actions Selection

flowchart TD
    T1["Create or edit topics"]
    A1["Assign standard actions"]
    A2["Assign custom actions"]
    U["User utterance"]
    P["Planner classifies topic"]
    S["Topic scope and instructions"]
    SEL["Select actions"]
    EXEC["Execute actions"]
    OUT["Return outputs"]

    T1 --> A1
    T1 --> A2
    U --> P
    P --> S
    S --> SEL
    SEL --> EXEC
    EXEC --> OUT

4) Custom Action Creation

flowchart TD
    N["New custom action"]
    R{"Select reference type"}
    AX["Invocable Apex class"]
    FL["Autolaunched flow"]
    PT["Prompt template"]
    IOP["Define inputs and outputs"]
    INS["Write instructions and examples"]
    SAVE["Save and assign to topics"]

    N --> R
    R -->|Apex| AX
    R -->|Flow| FL
    R -->|Prompt| PT
    AX --> IOP
    FL --> IOP
    PT --> IOP
    IOP --> INS
    INS --> SAVE

5) Testing Center Flow

flowchart TD
    T0["Prepare CSV test cases"]
    T1["Open Testing Center"]
    T2["Create test in sandbox"]
    T3["Upload CSV and run"]
    T4{"Results"}
    T5["Approve"]
    T6["Analyze and refine"]

    T0 --> T1
    T1 --> T2
    T2 --> T3
    T3 --> T4
    T4 -->|Passed| T5
    T4 -->|Failed| T6
    T6 --> T3

6) Deployment: Sandbox to Production

flowchart TD
    D0["Agent tested in sandbox"]
    D1{"Deployment method"}
    CS1["Create outbound change set"]
    CS2["Add agent metadata"]
    CS3["Upload to production"]
    CS4["Validate and deploy"]
    MD1["Retrieve with CLI"]
    MD2["Deploy to production"]
    ACT["Activate in Agent Builder"]
    COM["Notify and train users"]

    D0 --> D1
    D1 -->|Change set| CS1
    CS1 --> CS2
    CS2 --> CS3
    CS3 --> CS4
    D1 -->|Metadata API| MD1
    MD1 --> MD2
    CS4 --> ACT
    MD2 --> ACT
    ACT --> COM

7) Security and Permissions

flowchart TD
    A["Admin assigns permissions"]
    B{"User task"}
    P1["Create default agent permissions"]
    P2["Create other agent permissions"]
    P3["Use default agent permission set group"]
    C["Create agent user record"]
    L["Assign permission set with agent user license"]

    A --> B
    B -->|Create default agent| P1
    B -->|Create other agents| P2
    B -->|Use default agent| P3
    A --> C
    C --> L

8) Monitoring and Analytics Loop

flowchart TD
    M1["User interactions"]
    M2["Event logs"]
    M3["Data Cloud storage"]
    M4["Analytics dashboards"]
    M5["Utterance insights"]
    M6["Refine topics actions instructions"]

    M1 --> M2
    M2 --> M3
    M3 --> M4
    M4 --> M5
    M5 --> M6
    M6 --> M1

📚 Flashcards

Recent Notes

Graph View