Salesforce Agentforce Specialist: 5 Must-Know Concepts

The Salesforce Agentforce Specialist certification validates expertise in designing, configuring, and deploying AI-powered agents. These five concepts are the exam spine — agents, prompts, grounding, Builder, and integration.

1. AI Agents Fundamentals#

AI Agents make up about 35% of the exam — weigh study time accordingly.

Tip

Agentforce agents use context, tools, and grounding to take actions — not just fire static automation. Exam stems often hinge on that distinction.

Agent Types#

Salesforce provides different agent types optimized for specific use cases:

Agent Type	Purpose	Key Features
Service Agent	Customer service and support	Case handling, Knowledge integration, escalation flows
Sales Agent	Sales enablement and coaching	SDR activities, opportunity insights, sales coaching
Employee Agent	Internal employee assistance	HR queries, IT support, policy questions
Custom Agent	Specialized business needs	Fully customizable topics, actions, and grounding

Warning

A common exam trap: Using the wrong agent type for a scenario. Sales Agents are NOT for customer service, even if the end goal involves sales. Match the primary function to the agent type.

Topics and Actions#

Agents are configured with Topics (conversation categories) and Actions (specific capabilities):

Standard Topics — Pre-built conversation handlers for common scenarios
Custom Topics — Business-specific conversation flows you define
Actions — Discrete tasks an agent can perform (query data, update records, call APIs)

Agent User Security#

The Agent User is a special user type that:

Defines what data the agent can access
Controls which objects and fields are visible
Sets the security context for agent operations

AI Agents Check

A company wants to build an agent to help sales reps identify upsell opportunities based on customer purchase history. Which agent type is most appropriate?

2. Prompt Engineering Mastery#

Prompt Engineering accounts for 20% of the exam. This domain tests your ability to design effective prompts that produce accurate, grounded responses.

Prompt Templates#

Prompt templates are reusable structures that define how agents interact:

Component	Description	Example
Preamble	Sets context and agent persona	"You are a helpful customer service assistant..."
Flex Types	Dynamic variables for runtime context	`{case_description}`, `{customer_history}`
Instructions	Specific behavioral guidance	"Always verify account status before proceeding"
Output Format	Defines response structure	"Respond in bullet points with a summary"

Note

Flex Types are crucial for personalization. They allow prompts to dynamically include context like case details, customer history, or product information at runtime.

Grounding Techniques#

Grounding connects agent responses to your actual data, reducing hallucinations:

Knowledge Grounding — Link to Salesforce Knowledge articles
Data Cloud Grounding — Connect to indexed documents and data
Record Grounding — Reference specific Salesforce records

Best Practices vs Anti-Patterns#

Best Practice	Anti-Pattern
Use specific, measurable instructions	Vague directions like "be helpful"
Include output format requirements	Assuming the model will format correctly
Ground responses in indexed data	Allowing free-form responses without grounding
Test with edge cases	Only testing happy path scenarios
Use appropriate retriever types	Using keyword retriever for semantic queries

Prompt Engineering Quiz

A Service Agent frequently produces responses that mention products the company doesn't sell. What is the most effective solution?

3. Data Cloud & Grounding#

Data Cloud for Agentforce represents 20% of the exam. This domain covers how agents access and retrieve information.

Data Library Architecture#

The Data Library is where you ingest and index content for agent grounding:

Data Library
    └── Data Sources (PDFs, web pages, files)
            └── Chunking (splitting into segments)
                    └── Indexing (creating searchable embeddings)
                            └── Retrievers (finding relevant chunks)

Retriever Types#

Understanding when to use each retriever type is critical:

Retriever	How It Works	Best For
Keyword	Exact text matching	Structured data, product codes, specific terms
Vector	Semantic similarity using embeddings	Unstructured text, natural language queries
Hybrid	Combines keyword + vector	Mixed data types, best overall accuracy

Tip

For compliance and accuracy-critical applications, hybrid retrievers typically outperform single-type retrievers. They capture both exact matches AND semantic meaning.

Chunking and Indexing#

When ingesting documents:

Chunking — Splits large documents into manageable segments
Chunk Size — Affects context window utilization and retrieval precision
Overlap — Preserves context across chunk boundaries
Indexing — Creates vector embeddings for semantic search

Warning

Critical exam concept: When source documents are updated, vector embeddings must be regenerated. Stale indexes will return outdated information even if the source files are current.

Handling Unstructured Data#

For PDFs, emails, and other unstructured content:

Ingest into Data Library
Configure appropriate chunking strategy
Build vector index for semantic search
Optionally configure PII exclusion
Connect to agent via grounding configuration

Data Cloud Quiz

An agent needs to answer questions about product specifications stored in technical PDFs. Users ask questions using natural language, not exact product codes. Which retriever is most appropriate?

4. Development Lifecycle#

Development Lifecycle accounts for 20% of the exam. This domain covers testing, deployment, and monitoring of agents.

Testing Center#

The Agentforce Testing Center is your primary tool for validating agent behavior:

Capability	Description
Conversation Testing	Simulate user interactions with expected outcomes
Regression Testing	Verify changes don't break existing functionality
Escalation Testing	Validate handoff flows to human agents
Response Validation	Check responses against expected answers

Note

Always test in the Testing Center before deploying to production. Post-deployment monitoring cannot replace pre-deployment validation.

Deployment Best Practices#

The recommended deployment flow:

Development — Build and configure in a sandbox
Testing — Validate in Testing Center with comprehensive test cases
UAT — User acceptance testing with real scenarios
Production — Deploy with monitoring enabled

Monitoring and Governance#

Key monitoring areas:

Conversation Analytics — Track agent performance metrics
Escalation Rates — Monitor when agents hand off to humans
Response Quality — Review agent responses for accuracy
Trust Layer Logging — Audit trail for compliance

Einstein Trust Layer#

The Trust Layer provides:

Data Masking — Protects sensitive information in prompts
Audit Logging — Records all agent interactions
Toxicity Detection — Filters inappropriate responses
Grounding Validation — Ensures responses use provided context

Lifecycle Quiz

A team has built a complex agent with custom topics and Data Cloud grounding. Before production deployment, they need to validate responses and test escalation flows. What should they use?

5. Multi-Agent Interoperability#

Multi-Agent Interoperability is 5% of the exam, but expect nuanced questions about protocols and integration patterns.

Communication Protocols#

Agentforce supports several mechanisms for agent communication and external integration:

Protocol	Purpose	Use Case
A2A (Agent-to-Agent)	Inter-agent communication	Delegating tasks between agents with context
MCP (Model Context Protocol)	Context and permission management	Enforcing data scope during handoffs
Agent API	External system integration	Exposing agent capabilities to third-party systems

Tip

A2A is for Salesforce agents talking to each other. Agent API is for external systems calling into Agentforce. Don't confuse them on the exam.

Agent-to-Agent (A2A) Protocol#

A2A enables sophisticated multi-agent workflows:

Context Passing — Transfer conversation history between agents
Task Delegation — Hand off complex tasks to specialized agents
Status Reporting — Receive updates when delegated tasks complete
Security Boundaries — Maintain appropriate data access controls

Agent API#

The Agent API exposes agent capabilities externally:

RESTful endpoints for invoking agent capabilities
Authentication and rate limiting support
Structured request/response formats
Integration with external ticketing, CRM, or custom systems

Integration Patterns#

Common multi-agent patterns:

Escalation Pattern — Support agent delegates to specialist agent
Orchestration Pattern — Central agent coordinates multiple specialist agents
Hybrid Pattern — Agents work with both internal and external systems

Interoperability Quiz

A Customer Support Agent needs to delegate complex technical issues to a Technical Expert Agent, passing the full conversation context. Which protocol should be used?

Know your range before test day#

Agentforce scenarios reward precise product language. Sit a short diagnostic, then practice free on the domains that feel fragile.

Timed diagnostic with domain-level breakdown
Clear signal on what to drill next
Practice free; timed sittings and readiness range on Pro

Take the free Agentforce Specialist assessment →

Next step: Start practicing free on the Agentforce Specialist hub, or sit a short diagnostic first.