AgentForce Study Notes
1. Overview of the Certification
The Salesforce Agentforce Specialist Certification focuses on the configuration, deployment,
and optimization of AI-powered agents using the Salesforce platform. It validates expertise in
designing prompt templates, leveraging Data Cloud for dynamic data retrieval, enforcing data
security via the Einstein Trust Layer, and deploying agents in production.
2. Exam Coverage Areas
A. Prompt Builder & Templates
● Understand the different types of templates: Sales Email, Flex, Field Generation.
● Learn when to use each template type.
● Study the role of hyperparameters like Temperature, Presence Penalty, and Frequency
Penalty.
● Explore the Prompt Builder Scorecard for evaluating prompt effectiveness.
● Practice defining role-based prompts and grounding strategies.
B. Data Cloud & Grounding
● Learn how retrievers fetch relevant Salesforce data or documents.
● Study how to refine search scope using filters for accuracy.
● Understand supported grounding objects like Case and Knowledge.
● Examine how merge fields work with structured and external data sources.
C. Agent Deployment
● Know the deployment checklist: Flows, Apex classes, topics, and permissions.
● Understand Apex test coverage requirements (75% minimum).
● Learn about dependency packaging and version control.
● Get familiar with post-deployment activities like agent activation and channel
configuration.
D. Einstein Trust Layer
● Zero Data Retention Policy: Prevents third-party LLMs from storing or training on org
data.
● Prompt Defense: Guards against harmful or biased prompts.
● Data Masking: Obscures sensitive information in prompt execution.
● Audit Trails: Tracks and verifies data interactions for compliance.
E. Reasoning Engine & Copilot Planner
● Role of the Reasoning Engine in interpreting user intent.
● Understand how Planner Services select agent actions like Query Records.
● Learn the difference between standard and custom actions.
● Explore how Copilot integrates prompt responses with Salesforce data.
3. Key Topics to Master
● Designing and refining prompts based on use case scenarios.
● Deploying agents across channels (e.g., Messaging).
● Configuring Omni-Channel Flows for escalations.
● Working with Prompt Templates in dynamic use cases like Sales and Support.
● Enabling permissions like "Prompt Template User" for team access.
● Creating Close Plans to guide sales reps with AI-driven steps.
● Setting up Agentforce Testing Center for validation (no Einstein Requests consumed).
4. Study Patterns Observed
● Many questions are scenario-based and test judgment more than rote memorization.
● You'll often choose between similar-sounding options—clarity of concepts is key.
● Be familiar with standard naming conventions and configurations (e.g., grounding
objects, supported channels).
● Understand the role of every tool in the Agentforce ecosystem (Prompt Builder, Model
Builder, Data Cloud, Trust Layer).
5. Recommended Preparation Steps
1. Use Trailhead to prepare
2. Practice building prompts in a developer org and experiment with grounding options.
3. Use the notes below
6. Final Tips
● Treat this certification like a project—think through use cases end-to-end.
● Understand why certain configurations work better than others.
● Don’t rely on memorizing answers; master the architecture and flow of agent logic.
● Be hands-on—create a test agent in your dev org and explore actions like Query
Records, Close Plans, and Prompt Templates.
My Notes:
Prompt templates in Agentforce: Sales Email, Flex, Field Generation
Zero Data Retention Policy: Ensures no data is stored or used by third-party LLMs
Supported grounding objects: Case and Knowledge
Reasoning Engine: Maps user intent to agent actions
Apex deployment requirement: 75% test coverage
Prompt Builder: Designs and tests AI prompts
Permission to use templates: Prompt Template User
Permissions to create templates: Prompt Template Manager
Data Masking: Obscures sensitive fields in prompts
Evaluate prompt quality: Use the Prompt Builder Scorecard
Data retrieval tool: Retriever
SDR Agent deployment: Supported in Messaging
Field Generation prompts require API v59+
Planner Service: Selects agent actions based on input
Import any external models: Use Model Builder
Real-time ERP data grounding: External Object Record Merge Fields
Field must be linked to prompt template in layout
Prompt Defense: Protects against toxic or biased inputs
Copilot Builder: Builds assistant-like tools in Agentforce
Integrating external models: Use Model Builder
Flex vs Sales Email: Flex = Recommendations, Sales Email = Outreach
Knowledge answers use: Data Cloud-indexed articles
Filtered data retrieval: Query Records
Testing Center: Validates agents without consuming Einstein Requests
Unsupported related list merge: Activities and events
Secure Data Retrieval: Based on user’s active permissions and respects salesforce
security
For Prompt personalization: Use role-playing, tone, hyperparameters
Irrelevant AI responses are typically Caused by poor retriever filters
Data Cloud-triggered flow: Brings data to prompt templates
Managing templates: Requires Prompt Template Manager
Hyperparameter - Temperature: Controls randomness/creativity
Deployment readiness: Needs dependencies, test coverage, activation
Rich text fields: Not supported in Service AI Grounding
Agentforce lacks: Standard Delete Record action
Human escalation: Use Outbound Omni-Channel Flow
Adoption insights: Use AgentForce Analytics
After editing prompts: Review via Prompt Builder Scorecard
Action-selection logic: Handled by Reasoning Engine
AI safety check: Einstein Toxicity Score
Field Generation templates: Require API v59+
Flex templates: Used for product suggestions and dynamic responses
Grounding configuration: Select Case + Knowledge objects
Prompt Builder benefits: Delivers speed, accuracy, consistency
AI respects permissions: Enforced via CRUD & FLS
Filter outdated data: Apply filters in retriever settings
Improve LLM context: Add role-based instructions
Thank you! If you pass this exam, please recommend this channel to your friends and network.
