You Can Pass This Exam For Free
Choose Your Study Path
Limited Salesforce or AI experience. You need to build foundational knowledge in both the Salesforce platform and AI agent concepts before tackling Agentforce-specific topics.
Week 1Learn Salesforce platform basics: objects, fields, permission sets, profiles, Flows, and Apex triggers. Complete the Salesforce Admin Beginner trail on Trailhead
Week 2Study AI fundamentals: LLMs, prompt engineering, RAG, grounding, hallucinations, and the Einstein Trust Layer. Complete the Agentblazer learning path
Week 3Deep dive into Domain 1 (35% of exam): Atlas reasoning engine, agent types (Service, Sales SDR, Sales Coach, Employee), topics, instructions, actions, and Agent User security
Week 4Continue Domain 1: standard vs custom topics and actions, action types (Apex, Flow, Prompt Template, MuleSoft API), channel connections (Slack, Messaging, web), and deterministic behavior with filters and variables
Week 5Study Prompt Engineering (20%): Prompt Builder, template types (Sales Email, Field Generation, Flex), grounding techniques (merge fields, Flow merge fields, RAG), prompt template activation, and best practices for reducing hallucinations
Week 6Study Data Cloud for Agentforce (20%): Data Library types, unstructured data processing (chunking, indexing), search types (keyword, vector, hybrid), retrievers (individual, ensemble), and Data 360 concepts
Week 7Study Development Lifecycle (20%): Agentforce Testing Center, sandbox vs production deployment, monitoring adoption with Agentforce Analytics, Utterance Analysis, and Digital Wallet usage metering
Week 8Study Multi-Agent Interoperability (5%): Model Context Protocol (MCP), AgentExchange, agent-to-agent collaboration, Agent API, and the Einstein Trust Layer security boundary
Week 9Hands-on practice in a Developer Edition org: build an agent, configure topics and actions, create prompt templates, set up Data Library grounding, and test with the Testing Center
Week 10Practice questions across all domains, review explanations carefully, focus on Domain 1 which is 35% of the exam. Take full mock exams and review weak areas
Exam Overview
Format
60 scored multiple-choice and multiple-select questions plus up to 5 unscored pilot questions (65 total items), 105 minutes. No performance-based questions. Unscored questions are not identified and do not affect your score.
Scoring
Percentage-based scoring. Passing: 73%. No penalty for wrong answers — always guess if unsure.
Domains & Weights
- AI Agents35%
- Prompt Engineering20%
- Data Cloud for Agentforce20%
- Development Lifecycle20%
- Multi-Agent Interoperability5%
Registration
$200 USD. Available at Pearson VUE testing centers or online proctored from home. Exam fee is $200 USD. Retakes cost $100 USD.
Topic Priority Table
Not all topics are tested equally. Focus your study time on Tier 1 first, then Tier 2. Tier 3 topics rarely appear — just recognize what they do.
Domain 135% of exam
AI Agents
The heaviest domain at 35% — expect roughly 21 questions on this topic. Covers how agents work, the Atlas reasoning engine, agent types, topics and actions configuration, security through the Agent User concept, channel connections, and deterministic behavior controls. This is the make-or-break domain.
Key Topics
Atlas Reasoning EngineAgent BuilderTopics/Instructions/ActionsAgent UserPermission SetsChannel ConnectionsService AgentSales SDRSales CoachEmployee Agent
Must-Know Concepts
- Atlas reasoning engine: how it evaluates utterances, selects topics, plans action sequences, and delivers context-aware responses
- Agent types: Service Agent (customer-facing), Sales SDR (lead qualification, meeting booking), Sales Coach (personalized coaching with role-playing), Employee Agent (internal role-based collaborator)
- Topics define what jobs an agent can do. Instructions set decision-making boundaries. Actions are the specific tasks the agent performs
- Standard vs custom topics and actions: standard come pre-built with agent types; custom are created for unique business requirements
- Action types: invocable Apex, REST Apex, autolaunched Flows, prompt templates, and MuleSoft APIs — know when each is appropriate
- Agent User security: a special user type that agents run as, with permission sets controlling data and action access. Differs by agent type
- Channel connections: Slack, Messaging (SMS, WhatsApp), web chat, email — know which channels are available and how to configure them
- Deterministic behavior: using filters and variables to constrain agent behavior for predictable responses in specific scenarios
- Einstein Trust Layer integration: how agents operate within the trust boundary including data masking, toxicity detection, and audit logging
Common Exam Traps
Atlas reasoning engine SELECTS topics based on the user utterance — it does not execute actions directly. It plans the action sequence, then the actions execute
Agent User is NOT the same as the end user. The Agent User is the system user the agent runs as. The end user is the person interacting with the agent
Standard topics and actions vary by agent type. A standard topic for Service Agent is different from a standard topic for Sales SDR
Custom actions built on Apex require 75% test coverage, just like any Apex code in Salesforce. Do not forget this Salesforce platform requirement
Connecting an agent to a channel does not automatically grant permissions. Agent User permission sets must be configured separately for each channel context
Domain 220% of exam
Prompt Engineering
Covers Prompt Builder, prompt template types, grounding techniques, best practices for effective prompts, and template lifecycle management. You must know when to use Prompt Builder, how to select the right template type, and how to ground prompts in Salesforce data to reduce hallucinations.
Key Topics
Prompt BuilderSales Email TemplateField Generation TemplateFlex TemplateMerge FieldsFlow Merge FieldsRAG GroundingTemplate Activation
Must-Know Concepts
- Prompt Builder: the Salesforce tool for creating and managing reusable prompt templates. Know when it is appropriate to use Prompt Builder vs other approaches
- Template types: Sales Email (email draft generation), Field Generation (auto-fill record fields), Flex (general-purpose flexible templates) — know when each applies
- Grounding techniques: merge fields (direct record data), Flow merge fields (processed/calculated data), RAG from Data Cloud (semantic search across knowledge bases)
- Best practices: set output format and tone, include guardrails in prompts, ground prompts in relevant data, iterate and test prompts, reduce hallucinations through specific context
- Template activation: the process of making a prompt template available for use. Know the activation workflow and requirements
- User roles in Prompt Builder: who can create, edit, activate, and use prompt templates based on their Salesforce permissions
- Prompt template lifecycle: ideation, building, testing, activation, deployment, and observation — know each stage
Common Exam Traps
Field Generation templates are tied to a SPECIFIC OBJECT AND FIELD. They are not general-purpose like Flex templates
Grounding a prompt with a merge field pulls data at RUNTIME, not at template creation time. The data is dynamic based on the record context
Prompt Builder templates can be used as ACTIONS in Agent Builder. A prompt template action is one of the action types an agent can execute
Sales Email templates generate email content specifically. Do not confuse them with Flex templates that can generate any type of text output
Activating a prompt template is a separate step from creating it. An inactive template exists but cannot be used by agents or users
Domain 320% of exam
Data Cloud for Agentforce
Covers how Data Cloud (Data 360) provides grounding data to Agentforce agents. You must understand Data Library types, unstructured data processing through chunking and indexing, search index types (keyword, vector, hybrid), retrievers, and how data quality impacts agent response accuracy.
Key Topics
Data Cloud (Data 360)Data LibrarySearch IndexesRetrieversVector EmbeddingsChunkingIndexingKnowledge Base
Must-Know Concepts
- Data Library: the storage layer managing knowledge sources for agent grounding. Sources include Salesforce Knowledge articles, uploaded files (text, HTML, PDFs), and Data Cloud objects
- Chunking: breaking unstructured data into semantically appropriate chunks for processing. Chunk size and method affect retrieval quality
- Indexing: creating vector embeddings (numerical representations) from chunked data. Field selection during indexing directly impacts retrieval quality
- Search index types: keyword (exact lexical matching), vector (semantic similarity), hybrid (combines both). Know strengths and weaknesses of each. Hybrid and vector indexes can also be enriched with automatically extracted metadata, entities, and question chunks to improve RAG retrieval accuracy
- Retrievers: components that search and return relevant data. Individual retrievers query one index; ensemble retrievers combine results from multiple indexes
- Data 360 rebranding: Data Cloud was rebranded to Data 360 in late 2025. The exam may reference either name
- Unstructured data support: text documents, PDFs, HTML files, multimedia, chat logs, and customer feedback can all be processed
- Data quality impact: poor data quality, incorrect chunking, or misconfigured indexes directly reduce agent response accuracy and relevance
Common Exam Traps
Vector search understands MEANING but struggles with exact numbers and domain-specific terms. Keyword search handles exact terms well but does not understand context. Hybrid combines both strengths. Enriched indexes add extracted metadata and question chunks on top of vector or hybrid indexes — they are an enhancement, not a separate index type
Ensemble retrievers are NOT always better than individual retrievers. They add complexity and latency. Use them only when data spans multiple sources
Chunking is not automatic — field selection and chunk size must be configured. Poor chunking leads to irrelevant retrieval results
Data Library sources are different from Data Cloud data streams. Data Library is specifically for agent grounding, not general data ingestion
Data Cloud / Data 360 rebranding means the same technology. Do not treat them as different products on the exam
Domain 420% of exam
Development Lifecycle
Covers the end-to-end lifecycle of Agentforce agents from ideation through production monitoring. You must understand sandbox development, the Testing Center for validation, deployment processes, and post-deployment monitoring with Agentforce Analytics and Utterance Analysis.
Key Topics
Agentforce Testing CenterSandbox EnvironmentsDeployment PipelineAgentforce AnalyticsUtterance AnalysisDigital WalletAgentforce Observability
Must-Know Concepts
- Agent lifecycle stages: ideation, building, testing, deployment, and observation — know what happens at each stage and the tools used
- Agentforce Testing Center: batch testing of agent configurations, automated test case execution, performance measurement, and validation before production deployment
- Sandbox development: Data Cloud and Agentforce support sandbox environments for safe development and UAT. Einstein Trust Layer audit trail works in sandboxes
- Deployment: moving agent configurations from sandbox to production. Understand the deployment process and what artifacts need to be migrated
- Agentforce Analytics: post-deployment monitoring built on Data Cloud. Provides adoption metrics, accuracy insights, and usage patterns
- Utterance Analysis: monitoring tool that analyzes agent conversations to identify areas for improvement, new topic opportunities, and response quality issues
- Digital Wallet: consumption metering for Data Cloud and Agentforce usage. Provides visibility into AI resource consumption across development and production
- Agentforce Observability: comprehensive monitoring including live health, adoption analytics, consumption tracking, and performance dashboards
Common Exam Traps
Testing Center tests agent CONFIGURATIONS, not just individual prompts. It validates the entire agent behavior including topic selection and action execution
Sandbox and production are SEPARATE environments. Agent configurations must be explicitly deployed — they do not sync automatically
Utterance Analysis is a POST-DEPLOYMENT tool. It analyzes real conversations, not test cases. Do not confuse it with the Testing Center
Digital Wallet tracks CONSUMPTION costs, not agent performance metrics. Performance metrics come from Agentforce Analytics
The lifecycle is iterative: observation feeds back into ideation and building. Utterance Analysis insights should drive agent configuration improvements
Domain 55% of exam
Multi-Agent Interoperability
The smallest domain at 5% — expect roughly 3 questions. Covers Model Context Protocol (MCP) for connecting agents to external tools and systems, agent-to-agent collaboration, AgentExchange marketplace, and the Agent API for programmatic access. Small weight but do not skip it — 3 easy points can make the difference.
Key Topics
Model Context Protocol (MCP)AgentExchangeAgent APIAgent-to-Agent CollaborationMCP ServersClient-Server Architecture
Must-Know Concepts
- Model Context Protocol (MCP): an open standard originally from Anthropic for connecting AI agents to external tools, systems, and data through a standardized client-server architecture
- MCP in Salesforce: Agentforce uses MCP to consume external API assets and expose Salesforce data to the broader agent ecosystem. Native MCP client capabilities were added to Agentforce in 2025 with enterprise-grade policy enforcement (security, rate-limiting, and access controls). MCP actions execute within the Einstein Trust Layer boundary
- AgentExchange: curated marketplace of vetted MCP servers and pre-built agent components. Deploy external connections through Agent Builder with no code required
- Agent-to-agent collaboration: MCP enables agents from different systems to communicate and collaborate on tasks that span multiple platforms
- Agent API: programmatic interface for interacting with Agentforce agents from external applications and systems
Common Exam Traps
MCP is an OPEN STANDARD, not a Salesforce-proprietary protocol. It was originally developed by Anthropic and adopted by Salesforce
MCP actions execute WITHIN the Einstein Trust Layer boundary. PII is masked before anything reaches an LLM, even through MCP connections
AgentExchange provides VETTED servers, meaning they are curated for security and reliability. It is not an open marketplace where anyone can publish
Agent API is for EXTERNAL programmatic access. Internal agent-to-agent collaboration within Salesforce uses different mechanisms
Concepts You Must Not Confuse
These pairs appear on nearly every exam. Learn the difference and you'll avoid the most common traps.
Top Mistakes to Avoid
Confusing the Agent User (system user the agent runs as) with the end user (person interacting with the agent) — they have different permission models
Mixing up standard topics/actions (pre-built per agent type) with custom topics/actions (user-created for unique requirements)
Thinking merge fields and Flow merge fields are interchangeable — merge fields pull raw record data while Flow merge fields pull processed/calculated data
Confusing keyword search (exact matching, good with numbers) with vector search (semantic meaning, struggles with exact terms) — know when to use hybrid
Not understanding the difference between individual retrievers (single index) and ensemble retrievers (multiple indexes) and when each is appropriate
Treating the Agentforce Testing Center (pre-deployment validation) and Utterance Analysis (post-deployment conversation analysis) as the same tool
Assuming Data Cloud and Data 360 are different products — Data Cloud was rebranded to Data 360 in late 2025; they are the same technology
Forgetting that Apex-based custom actions require 75% test coverage, the same platform requirement as any Salesforce Apex code
Confusing Field Generation templates (auto-fill specific record fields) with Flex templates (general-purpose, not tied to specific fields)
Thinking MCP is Salesforce-proprietary — it is an open standard originally from Anthropic, adopted by Salesforce for agent interoperability
Exam-Ready Checklist
Can explain all 5 exam domains and their relative weights (35%, 20%, 20%, 20%, 5%)
Know how the Atlas reasoning engine processes utterances: topic selection, action planning, and execution
Can distinguish between all agent types (Service, Sales SDR, Sales Coach, Employee) and select the right one for business scenarios
Understand the Agent User concept and how permission sets control agent data and action access
Can explain all three grounding techniques (merge fields, Flow merge fields, RAG) and when each is appropriate
Know all Prompt Builder template types (Sales Email, Field Generation, Flex) and their specific use cases
Can distinguish between keyword, vector, and hybrid search types with their strengths and weaknesses
Understand individual vs ensemble retrievers and when to use each based on data source distribution
Know the complete agent lifecycle: ideation, building, testing (Testing Center), deployment, and observation (Analytics, Utterance Analysis)
Can explain the Einstein Trust Layer components: data masking, toxicity detection, prompt defense, dynamic grounding, and audit trail
Understand MCP architecture and how Salesforce implements it for multi-agent interoperability through AgentExchange
Know the difference between sandbox development/testing and production deployment, and what artifacts must be migrated
Completed hands-on practice in a Developer Edition org: built an agent, created prompt templates, configured Data Library grounding
Scored 75%+ on at least two full mock exams (73% passing score on the real exam)
Recommended Resources
Free & Official Resources
Paid Courses & Practice Exams
These are recommended if you prefer a structured learning path. They can save time but are not required to pass.