Agentforce Specialist (AI-201): 5 Practice Tests 2026
300 questions on Agent Script, Data 360, Trust Layer, Testing Center and MCP, with an explanation on every option
You have read the exam guide, you have finished the Trailhead badges, and you still cannot answer the only question that matters: are you ready to book this exam? These five practice tests are built to answer it.
Five timed tests of 60 questions each, 300 in total, matching the length of the real Agentforce Specialist exam. Every test carries the same section weighting as the current exam guide, which aligns to the Spring '26 release: AI Agents 35%, Prompt Engineering 20%, Data 360 Fundamentals 20%, Testing, Deployment and Maintenance 10%, Governance and Observability 10%, and Multi-Agent Orchestration 5%. Because the weighting is identical across all five, your score by section adds up to a real picture of where you stand.
That currency matters more here than in most certifications. This exam was renamed from Salesforce Certified AI Specialist in March 2025, and the outline was rewritten again after that. Data Cloud is now Data 360, agent topics are now subagents, and Agent Script with hybrid reasoning has become the heaviest part of the exam. Practice material written for the older outline still asks about Einstein Copilot and about Service Cloud and Sales Cloud sections that no longer exist. Preparing on it costs you the attempt, not just the study time.
The five tests rise in difficulty while the clock tightens. Test 1 is groundwork: one concept per question, short stems. Test 2 puts you in short scenarios where neighbouring features are easy to confuse. Test 3 gives you two defensible answers and one constraint that decides between them. Test 4 goes under the hood, into Agent Script syntax, service limits, configuration fields and the order of operations. Test 5 is a full-length simulation on the real exam's clock.
Every answer option has its own explanation, saying why it is right or, if it is wrong, what that option actually is. Each question closes with an overall rationale and a link to the Salesforce documentation behind it, so a wrong answer teaches you something instead of costing you a point. Pass marks are 80% for tests 1 to 4 and 85% for test 5, deliberately above the real exam, because practising at your desk is easier than sitting in a test centre.
These are practice exams, not a training course, and no question here comes from a real exam. Used alongside the official Trailhead material, they will tell you what you still do not know while you can still do something about it.
Groundwork: Agents, Prompts, and Data 360 Foundations
60 questions · 100 minutes · pass mark 80%
Start here, and take this test before you revise anything. It is a baseline measurement, not a verdict.
Sixty questions covering all six sections of the exam guide in their official proportions: AI Agents, Prompt Engineering, Data 360 Fundamentals, Testing, Deployment and Maintenance, Governance and Observability, and Multi-Agent Orchestration. The questions are direct: one concept each, short stems, no layered scenarios yet. Agent types and their security context, the Agent Script building blocks, the three lifecycle blocks, prompt template types, grounding methods, Trust Layer components, chunking and indexing, retriever types, the three Testing Center evaluations, the three Observability components, and the purpose of MCP and A2A.
You have 100 minutes and the pass mark is 80%, which is 48 of 60. That bar is higher than the real exam on purpose: answering from a chair at home is easier than answering in a proctored test centre, and the margin is your buffer. Seven questions ask for more than one answer and say so in the stem.
Every answer option carries its own explanation, and each question closes with a rationale and a link to the Salesforce documentation behind it.
In Practice: Choosing the Right Building Block
60 questions · 95 minutes · pass mark 80%
The second test moves from definitions into short practical scenarios. Every question puts you in a situation and asks which building block fits.
The theme is the pairs that are easy to confuse. Employee agent or Service agent for this use case. A Flow action, an Apex action, a prompt template action, or a standard action you customise. Messaging Enhanced Chat, Service Email, Partner Telephony or Bring Your Own Channel. Field Generation, Flex, Sales Emails, Record Summary or Knowledge Answers. Record merge fields or a retriever. An individual, ensemble or web retriever. Which of the three Testing Center evaluations actually failed. Which Observability component answers the question you are asking.
Sixty questions, 100 seconds each, 95 minutes in total, pass mark 80%, which is 48 of 60. Eight questions ask for more than one answer.
The section weighting is the same as in every other test in this practice set, so comparing your score by domain across attempts tells you something real.
Trade-offs: When Two Answers Look Right
60 questions · 100 minutes · pass mark 80%
The third test is where the exam usually gets people. Two of the answers are defensible, and one constraint in the question decides between them.
You will be told that the agent must never surface a record the requester cannot already open, or that the reasoning has to be reproducible for an audit, or that there is no user interface at all, or that the content is unstructured, or that the existing bot has to keep running through the migration, or that a conversation must never be routed to a service representative and an AI agent at the same time. Read the constraint before you read the options; it is the whole question.
Sixty questions across the same six sections in the same proportions, 100 minutes, pass mark 80%, which is 48 of 60. Eight questions ask for more than one answer.
Expect this test to score lower than the previous two. That is the design, and the explanations are where the value is.
Under the Hood: Script, Limits, and Configuration
60 questions · 80 minutes · pass mark 80%
The fourth test goes under the hood, into the detail an experienced specialist is expected to have at hand.
Agent Script syntax: the shape of an action definition, the target reference, deterministic execution versus offering an action to the model as a tool, action chaining, and the naming rules. Variables: which types each kind supports, which ones can carry a default, which ones the agent may change, and how visibility affects API access. Data 360: the file size limits of a data library, the fields you pick when creating a search index, the chunk and index objects it produces, and how a retriever is configured. Testing: the order of the test suite workflow and the lifecycle of a custom scorer. Observability: what is captured and what has to be enabled first.
Sixty questions, 80 minutes, pass mark 80%, which is 48 of 60. Nine questions ask for more than one answer. The stems are shorter here than in the previous test but the clock is tighter, because this material is either known or it is not.
Exam Day: Full-Length 60-Question Simulation
60 questions · 105 minutes · pass mark 85%
This is the readiness gate. Sit it once, in one uninterrupted session, at the full time limit, and treat the result as your answer to the question of whether to book the exam.
Sixty questions in 105 minutes, the same length and the same clock as the real Agentforce Specialist exam. The stems are the longest in this practice set and most of them carry more than one constraint, mixing sections the way the real exam does: an agent type decision that also depends on the channel and the escalation path, or a grounding design that depends on the source type, the index type and the retriever type at once.
The pass mark is 85%, which is 51 of 60. That is deliberately above the real exam's published bar. Clearing 85% here in one sitting, without pausing and without looking anything up, is a meaningful signal. Clearing it by 2% after three attempts is not.
If you come in under the line, go back to your weakest domain, reread the explanations rather than only the questions, and retake this test cold after a few days.
What you will practice
- Test yourself against all six sections of the current Agentforce Specialist exam guide, using the same domain weighting as the real 60-question exam.
- Tell neighbouring Agentforce features apart: Employee versus Service agents, merge fields versus retrievers, vector versus hybrid search indexes.
- Read Agent Script with confidence: the three lifecycle blocks, deterministic actions versus LLM tools, and custom versus linked variables.
- Learn from every question through an explanation on each individual answer option plus an overall rationale that links to Salesforce documentation.
- Build exam stamina across five tests that rise from single-concept recall to full-length, multi-constraint scenario simulation.
- See your score per exam section after every attempt, so you know which domain to revise before you book the real exam.
Exam domains covered
Who is this for
These practice tests are for Salesforce administrators, developers and architects who are preparing to sit the Salesforce Certified Agentforce Specialist exam, and for anyone who already configures agents, prompt templates or Data 360 grounding and wants an honest measurement of how ready they are. They also suit consultants and solution designers who have to defend an Agentforce design in front of a client, and career changers who have finished the Trailhead material and now need question practice rather than more reading.
There is one thing you should know before you buy. This certification used to be called Salesforce Certified AI Specialist; it was renamed to Agentforce Specialist in March 2025, and existing holders were converted automatically. The exam outline has since been rewritten again and now aligns to the Spring '26 release, with six sections: AI Agents, Prompt Engineering, Data 360 Fundamentals, Testing, Deployment and Maintenance, Governance and Observability, and Multi-Agent Orchestration. Agent Script, hybrid reasoning, Testing Center, Agent API, MCP and A2A are all in scope now. Practice material written for the older outline still tests Data Cloud, topics and Einstein Copilot, and it will prepare you for an exam that no longer exists. These tests follow the current exam guide.
What this is: 300 exam-style questions with a written explanation on every answer option, a rationale for each question, and a source link to the official documentation. What this is not: a course, a set of leaked exam questions, or a guarantee. Nothing here comes from a real exam. If you have never built an agent or a prompt template, work through the free official material first and use these tests to find your gaps.
Requirements
- No prior Agentforce experience is needed to start, although the exam itself assumes roughly a year of Salesforce platform configuration and standard objects, including Data 360.
- These are practice exams, not a training course. Use them alongside the free official material on Trailhead and Salesforce Help, not instead of it.
- A general idea of what a large language model does is enough. No coding is tested: Apex and Python syntax, model tuning and transformer internals are all out of scope.
Free exam guide
Not sure what to expect? Read how the exam is built, what it covers and how to prepare.