Salesforce Certified Agentforce Specialist Study Guide

Domain 1: AI Agents

Key concepts you must know · 240 practice questions

The Agentforce building blocks - agent, topics, instructions, and actions - and how Agent Builder assembles them.
How the Atlas Reasoning Engine interprets an utterance, classifies it to a topic, and selects and sequences actions.
Standard versus custom topics, and standard versus custom actions (Flow, Apex, Prompt Template, and CRM actions).
Managing deterministic agent behavior with filters and variables (input, output, and conversation variables).
Agentforce security: the Agent User run-as identity, the Agentforce User license, and enforcement of object, field, and record security during execution.
Agent types (Employee, Service, and Sales agents), channels (Experience Cloud, Messaging, Slack, web, voice), human escalation, and the Einstein Trust Layer (masking, toxicity detection, zero retention, and audit).

Domain 2: Prompt Engineering

Key concepts you must know · 130 practice questions

When to use Prompt Builder versus Flow, Apex, or a standard action, and the roles and permissions to create, manage, and execute prompt templates.
Prompt template types (Field Generation and Flex), merge fields, and related-list data.
Grounding techniques: record and merge-field grounding, related-list grounding, Flow and Apex grounding, and Data Cloud retriever grounding.
The create, activate, and test (execute) lifecycle of a prompt template.
Prompt-writing best practices: clarity, output format, tone and persona, sufficient context, and guardrails against hallucination.
How the Einstein Trust Layer applies inside prompt templates, and where templates are invoked (standalone, in Flow, or as an agent action).

Domain 3: Data 360 Fundamentals

Key concepts you must know · 144 practice questions

The Agentforce Data Library and its content: structured CRM objects and DMOs, and unstructured files, PDFs, Knowledge, and websites.
Data streams that ingest source data into DLOs and DMOs, and data graphs that model related data for grounding.
Improving accuracy on unstructured data with chunking and indexing.
Retrievers (individual versus ensemble) and search index types: keyword, vector (semantic), and hybrid.
How retrieval-augmented generation grounds agent and prompt responses in Data Cloud content.
Data governance so agents only return content the running user is allowed to see (Data Cloud is being rebranded as Data 360).

Domain 4: Testing, Deployment, and Maintenance

Key concepts you must know · 105 practice questions

Testing agents in the Agentforce Testing Center: test sets of sample utterances, expected topic and action, and pass or fail criteria.
Running regression tests before changes go live.
Deploying agents, topics, actions, and prompt templates from sandbox to production with change sets and the Metadata API.
Handling environment-specific credentials and named connections during deployment.
Monitoring adoption after deployment: conversation analytics, topic performance, and containment versus escalation rate.
Iterating on live agents based on real usage and user feedback.

Domain 5: Multi-Agent Orchestration

Key concepts you must know · 48 practice questions

The Model Context Protocol (MCP) for securely sharing context and tools with external models or systems.
The agent-to-agent (A2A) protocol that lets multiple agents collaborate on a task.
Agent API use cases for invoking an agent programmatically from an external application.
When interoperability across agents or vendors is the right architecture.

Domain 6: Governance and Observability

Key concepts you must know · 67 practice questions

Monitoring a running agent: session and intent views, session tracing, and the error status and error types a trace exposes (system problems, API failures, and agent logic).
Agent analytics: the Service Agent and Employee Agent dashboards, the Effectiveness tab and Task Resolution, and service-level signals such as response time and agent response latency.
How agent analytics is built on Data 360, and how its semantic data model can be extended per agent type.
Agent optimization: using intents and scorers to find where an agent is failing, then re-testing the change.
Governing connections: policies that protect calls to external APIs and MCP servers, with usage limits and tool restrictions applied by selection or by matching criteria.
The security context an agent runs under, its default user, and the permissions needed to activate and manage an agent.

Salesforce Certified Agentforce Specialist exam tips

AI Agents is by far the largest area at 35 percent. Know the agent building blocks (agent, topics, instructions, actions), how the Atlas Reasoning Engine selects a topic and sequences actions, standard versus custom topics and actions, and the Agent User security model.
The Einstein Trust Layer is the most cross-cutting concept on the exam. Be able to describe PII masking before data reaches the LLM, toxicity detection, zero data retention, and audit logging, and recognize where it sits in both agent and prompt-template flows.
For Prompt Engineering, master the grounding techniques (record, related list, Flow, Apex, and Data Cloud retriever) and the create-activate-test lifecycle, and know when Prompt Builder is the right tool versus a Flow or Apex action.
For Data Cloud, distinguish the search index types - keyword, vector, and hybrid - and understand retrievers, chunking and indexing of unstructured data, and how RAG grounds responses. Remember Data Cloud is being renamed Data 360.
Do not skip the smaller domains: Testing, Deployment, and Maintenance (10%) centers on the Agentforce Testing Center and sandbox-to-production promotion, Governance and Observability (10%) covers monitoring running agents and agent analytics, and Multi-Agent Orchestration (5%) covers MCP, the agent-to-agent protocol, and the Agent API.

Study guide FAQ

How many questions are on the exam and what is the passing score?

There are 60 scored multiple-choice and multiple-select questions in 105 minutes, and you need 72 percent to pass. The exam is proctored online or at a test center. The standard fee is 200 USD, though Salesforce has offered it free during promotional windows, so confirm the price at registration.

Is this the same as the old AI Specialist certification?

Yes. Salesforce renamed the AI Specialist credential to Certified Agentforce Specialist in March 2025, and existing holders were converted automatically. The separate, more foundational AI Associate certification was retired in early 2026 and is a different exam.

Do I need to write code?

No heavy coding. This is a configuration and solution-design exam focused on building agents with Agentforce, prompt templates with Prompt Builder, and grounding with Data Cloud. You should recognize when a Flow or Apex action is the right agent action, but you are not asked to write large programs.

What changed in the current version?

The Spring '26 blueprint has six sections. AI Agents is the largest at 35%, and the exam separates Testing, Deployment, and Maintenance (10%) from Governance and Observability (10%), which covers monitoring running agents and agent analytics. Multi-Agent Orchestration (5%) covers MCP, the agent-to-agent protocol, and the Agent API. Older guides showed a different split centered on Prompt Engineering and Agentforce Concepts.

Is CertGrid practice official Salesforce material?

No. CertGrid is an independent practice platform and is not affiliated with or endorsed by Salesforce. These questions are original and written to the published exam objectives. Our five domains do not yet match the six in Salesforce's Spring '26 outline - the note under Official exam sources below says exactly where they differ. Always confirm the current exam guide on Trailhead before your exam.

Official exam sources

A current vendor blueprint exists and the domains on this page do not yet match it. They follow the structure our question bank is built to; the note below says exactly where the two differ, so the gap can be seen rather than guessed at.

Salesforce - Certified Agentforce Specialist Exam GuideBlueprint revised Spring '26 release · link and content verified 8 September 202660 multiple-choice questions plus up to five unscored, 105 minutes, registration USD 200, and "Prerequisite: None". The passing score is 72%, and OUR PAGE SAID 73% - in prose and in its FAQPage structured data - so this source corrected a pass mark we were attributing to Salesforce. Read through a rendered browser: help.salesforce.com is a JavaScript application and a plain fetch returns only its loading shell.

This page and the current vendor source differ. The pass mark is fixed; the blueprint is not, and it has not been touched. Salesforce's Spring '26 outline totals 100% across SIX sections - AI Agents 35%, Prompt Engineering 20%, Data 360 Fundamentals 20%, Testing, Deployment, and Maintenance 10%, Governance and Observability 10%, Multi-Agent Orchestration 5%. Our five domains are AI Agents 35, Prompt Engineering 20, Data Cloud 20, Dev Lifecycle 20, Multi-Agent Interop 5: our domain 4 carries both of Salesforce's 10% sections at 20%, Governance and Observability has no counterpart in the bank, and Data Cloud is the product's former name. Splitting a domain means having questions to put in it, so this is recorded as an exam-alignment task rather than a weighting edit.

Related Salesforce resources

Test yourself: 734 Salesforce Certified Agentforce Specialist practice questions