Salesforce Certified Agentforce Specialist (AI-201) Practice Exam
Cover all six domains on Salesforce’s current Agentforce Specialist outline — AI Agents, Prompt Engineering, Data 360 Fundamentals, Testing and Deployment, Governance and Observability, and Multi-Agent Orchestration — with objective-mapped questions, instant feedback in Learn mode, and a full timed simulation in Exam mode.
Start 24-hour free trial →The blueprint changed — check what your study material targets. Salesforce has revised the Agentforce Specialist outline, and questions now align to the Spring ’26 release. Material written against the earlier five-domain structure is out of date in several concrete ways:
- The exam now has six domains, not five.
- Data Cloud for Agentforce is now Data 360 Fundamentals.
- The old Development Lifecycle domain has been split into Testing, Deployment & Maintenance (10%) and Governance & Observability (10%).
- Multi-Agent Orchestration is a brand-new 5% domain covering MCP, A2A, and Single-Agent (SOMA) architecture.
- The passing score is 72%.
The terminology moved too. Agentforce Builder (next-generation authoring) and Agent Script now underpin agent authoring, with hybrid reasoning replacing the purely probabilistic loop of the earlier release. The page below reflects the current outline.
Agentforce Specialist exam at a glance
- Vendor
- Salesforce
- Exam code
- AI-201
- Credential
- Salesforce Certified Agentforce Specialist
- Blueprint
- Current exam guide — six domains weighted 35 / 20 / 20 / 10 / 10 / 5
- Release alignment
- Exam questions align to the Spring ’26 release
- Format
- 60 multiple-choice questions, plus up to five unscored questions
- Duration
- 105 minutes
- Passing score
- 72%
- Registration fee
- USD 200 plus applicable taxes
- Retake fee
- USD 100 plus applicable taxes
- Prerequisites
- None. Salesforce expects roughly 1 year of platform configuration experience including Data 360, plus hands-on Agent Builder and Prompt Builder work.
- Recommended credentials
- Certified Platform Administrator or Certified Platform App Builder (recommended, not required)
- Delivery
- Proctored onsite at a testing centre or in an online proctored environment
- Languages
- English, French, and Japanese
- Maintenance
- Complete the certification maintenance module on Trailhead once a year, or the credential expires
Source: Salesforce — Certified Agentforce Specialist Exam Guide. Verify current details with Salesforce before scheduling; Agentforce changes with each release.
About the Salesforce Certified Agentforce Specialist certification
Agentforce Specialist is Salesforce’s core “you can build and run production AI agents” credential. Salesforce frames it for individuals who design and implement prompt templates and agents, validating the ability to build agents that reason and act across Salesforce and external channels. The stated audience is Administrators, Developers, and Architects who configure Prompt Builder for grounding, use Data 360 for retrieval, and manage the agent lifecycle from testing to production.
The exam guide is unusually clear about what is not expected, and that is worth reading before you over-prepare. You are not expected to configure or fine-tune large language models, know Apex or Python basics, hold expertise in external AI tools, understand transformer architecture, perform ROI calculations, or have experience with Marketing Cloud, Heroku, Commerce Cloud, MuleSoft, Tableau, Slack, Quip, or Industry Clouds. This is a configuration and design exam, not a machine-learning exam.
What has changed most is the authoring model. Salesforce now expects candidates to engineer agents using next-generation authoring in Agentforce Builder, understand hybrid reasoning and how Agent Script functions in both Canvas and Script view, and orchestrate systems using open multi-agent protocols including MCP and A2A. Every PowerKram question maps to one of the six current domains and cites the Salesforce documentation it was derived from. For the wider Salesforce landscape, see our definitive guide to Salesforce certifications.
Agentforce Specialist (AI-201) exam domains and weights
Salesforce publishes six weighted domains in the current exam guide. The weights sum to 100%. AI Agents dominates at 35% — more than a third of the exam — and together with Prompt Engineering and Data 360 Fundamentals accounts for three-quarters of the total.
How an agent works and the building blocks of agent script; the components and benefits of hybrid reasoning, including how Agent Script functions in Canvas and Script view; managing deterministic behaviour with filters, variables, and template expressions; selecting and configuring standard and custom topics and Agent actions; connecting agents to channels such as digital experience, email, voice, and Slack; the security context an agent actually runs in and how that affects action execution; choosing between an Employee and a Service agent; and knowing when to use the Agent API.
Identifying when Prompt Builder is appropriate; access controls governing prompt templates; considerations for template types such as field generation and flex; selecting the right grounding technique for a scenario; creating, activating, and executing templates; best practices for writing effective prompts; the security and privacy features of the Trust Layer; and managing or preventing access to specific models.
Considerations for the Agentforce Data Library and its concepts, and the foundational Data 360 concepts that make grounding work — chunking, indexing, and retrievers. Note that Data Cloud is now Data 360; the objective was renamed accordingly.
Testing an agent using Testing Center and understanding how its evaluations work; and the considerations for deploying both an agent and a prompt template from sandbox to production.
The process for managing and monitoring agents in production, plus agent analytics and agent optimization. This is a distinct domain in the current outline rather than part of a combined lifecycle objective.
Determining whether a Single-Agent (SOMA) architecture is appropriate for scalability and control, and explaining the purpose of existing open standard multi-agent protocols such as Model Context Protocol (MCP) and Agent-to-Agent (A2A) communication. This domain is new to the current outline.
Source: Salesforce — Certified Agentforce Specialist Exam Guide, Exam Outline. Weightings and domain names circulating elsewhere often reflect the earlier five-domain structure.
Who the Agentforce Specialist exam is for
Salesforce describes a candidate with roughly one year of experience with platform configuration and standard objects including Data 360, hands-on experience building agents in Agent Builder and prompt templates in Prompt Builder, and familiarity with Testing Center and sandbox-to-production deployment patterns.
- Salesforce administrators extending into AI — the configuration skills transfer directly, and the recommended Platform Administrator credential is the natural foundation.
- Developers and architects designing agents that act across systems, particularly now that Agent Script, the Agent API, and multi-agent protocols are on the blueprint.
- Consultants delivering Agentforce projects who need a credential clients recognise for the platform Salesforce is building its strategy around.
- Data 360 practitioners whose grounding, chunking, and retriever work sits directly under a 20% domain.
If you are newer to the platform, Certified Platform Administrator is the recommended starting point. If your work is heavier on the data side, Certified Data Cloud Consultant pairs closely with the Data 360 material here. To see where AI and agent work leads in terms of roles and progression, review the AI engineering career paths this certification supports.
What this Agentforce Specialist practice exam delivers
Learn mode
Get the correct answer, why each distractor fails, and a direct link to the Salesforce documentation behind it — immediately after each question. Most valuable across AI Agents, where the security context an agent runs in decides several answers.
Exam mode
Sixty questions on a 105-minute timer — the real AI-201 format. Build pacing on the scenario-based questions that dominate this exam.
Source-linked explanations
Every answer cites the Salesforce document it derives from — the exam guide, the Agentforce Builder and Agent Script documentation, and the Salesforce Architect hybrid reasoning guide — so you can verify rather than memorise.
Score by current domain
Results break down across the six current domains including the new Multi-Agent Orchestration objective, so your weak area is a specific reading list rather than a guess.
Sample Agentforce Specialist practice questions
Ten free questions spread across the six current domains, each with a full explanation and a source link to the Salesforce documentation it was built from. The complete bank is available with the 24-hour trial.
A team needs an agent to follow a guaranteed sequence of steps for a compliance workflow, while still handling free-form customer language naturally. Which Agentforce capability addresses both requirements?
- Increasing the temperature setting on the underlying model
- Hybrid reasoning, separating deterministic execution from LLM reasoning
- Adding more detail to a single long system prompt
- Disabling the Atlas Reasoning Engine for that agent
Show answer & explanation
Correct: B — hybrid reasoning. Agentforce's hybrid model separates deterministic execution from LLM reasoning, so the agent follows a precisely defined structure for every workflow execution while still using the LLM where judgement and natural language understanding are genuinely needed. This is exactly the combination the scenario describes.
Why not the others: Temperature (A) adjusts output randomness and does nothing to guarantee a sequence. A longer system prompt (C) is the approach hybrid reasoning replaces — it still leaves execution to probabilistic interpretation. The Atlas Reasoning Engine (D) consumes the compiled agent specification; disabling it is not a configuration option.
Source: Salesforce Architect — Hybrid reasoning with Agentforce Builder and Agent Script → Further reading: PowerKram — Salesforce certification definitive guide →In Agentforce Builder, a developer wants to review the underlying instructions that drive an agent's behaviour rather than the summarised visual blocks. Which view should they use?
- Canvas view, which shows the raw Agent Script
- The legacy Setup experience, which is the only place script is visible
- Script view, which shows the Agent Script itself
- Testing Center, which renders the script during evaluation runs
Show answer & explanation
Correct: C — Script view. In Agentforce Builder you work either in a visual Canvas view or directly in Script view, editing the Agent Script that defines the agent's behaviour. The current outline expects candidates to explain how Agent Script functions in both views.
Why not the others: Canvas view (A) is the visual layer that summarises the script into expandable blocks, not the raw script. The legacy Setup experience (B) remains available but is no longer the primary authoring path, and Builder is the recommended environment. Testing Center (D) evaluates agent behaviour rather than serving as an authoring view.
Source: Salesforce — New Agentforce Builder with Agent Script →An agent must retrieve records on behalf of a user, but returns fewer results than the user expects to see. Which consideration most likely explains the behaviour?
- The agent runs without any security context, so all records are hidden by default
- The security context the agent runs in constrains which records its actions can access
- Prompt templates always bypass sharing rules and return system-level data
- The Agent API disables record access entirely
Show answer & explanation
Correct: B — the agent's security context constrains action execution. The exam outline names explaining the security context in which the agent is actually running, and how it impacts agent action execution. Agents operate under a defined user context, so record visibility follows that user's access rather than being unlimited.
Why not the others: Agents do run in a security context (A) — the point is which one. Prompt templates do not bypass sharing to return system-level data (C); grounding respects access controls. The Agent API (D) is an interface for interacting with agents, not a mechanism that disables record access.
Source: Salesforce — Exam guide, AI Agents security context → Further reading: PowerKram — AI engineer role and career path →An organisation wants an agent that helps internal staff complete HR and IT requests inside Slack, rather than serving external customers. Which agent type fits?
- A Service agent
- A prompt template configured as flex type
- A retriever configured for hybrid search
- An Employee agent
Show answer & explanation
Correct: D — an Employee agent. The current outline asks candidates to identify, given a scenario, when to use an Employee or a Service agent. Internal staff assistance across functions such as HR and IT is the Employee agent use case; Slack is one of the named channels agents connect to.
Why not the others: A Service agent (A) is oriented toward customer-facing service interactions. A flex prompt template (B) is a prompt type, not an agent type. A retriever (C) is a Data 360 grounding component.
Source: Salesforce — Exam guide, agent types and channels →A compliance team requires that certain large language models must not be usable by any agent or prompt template in the org. Which capability addresses this?
- Lowering the token limit on those models
- Deactivating every prompt template that references them
- Managing and preventing specific models from being accessed
- Switching all templates to field generation type
Show answer & explanation
Correct: C — managing and preventing specific model access. The Prompt Engineering objective explicitly names explaining how to manage and prevent specific models from being accessed, alongside the security and privacy features of the Trust Layer. It is a governance control rather than a per-template workaround.
Why not the others: Token limits (A) constrain output length, not model availability. Deactivating templates (B) addresses today's templates but nothing created tomorrow. Template type (D) governs how a prompt is used, not which model backs it.
Source: Salesforce — Exam guide, Trust Layer and model access →A prompt template must populate a single field on a record with generated text, using values from that record. Which template type is designed for this?
- Field generation
- Flex
- Record summary only
- Retriever template
Show answer & explanation
Correct: A — field generation. The outline names field generation and flex as the template types candidates must weigh. Field generation is purpose-built for producing content that populates a specific field, grounded in the record it runs against.
Why not the others: Flex (B) is the flexible, general-purpose type used when the output is not bound to a single field. There is no separate "record summary only" template type (C). A retriever (D) belongs to Data 360 grounding rather than being a prompt template type.
Source: Salesforce — Exam guide, prompt template types → Further reading: PowerKram — Certified Data Cloud Consultant exam →An agent grounds its answers in a large library of PDF policy documents. Which pair of Data 360 concepts makes that unstructured content usable for retrieval?
- Chunking and indexing
- Deployment and observability
- Filters and variables
- Canvas and Script view
Show answer & explanation
Correct: A — chunking and indexing. The Data 360 Fundamentals objective names chunking, indexing, and retrievers as the foundational concepts. Long documents are split into chunks and indexed so a retriever can surface the relevant passages at query time rather than the whole file.
Why not the others: Deployment and observability (B) belong to two different domains entirely. Filters and variables (C) manage deterministic agent behaviour. Canvas and Script view (D) are Agentforce Builder authoring views.
Source: Salesforce — Exam guide, Data 360 Fundamentals →Before promoting an agent to production, a team wants to run it against a defined set of scenarios and score the results systematically. Which tool is intended for this?
- Prompt Builder preview
- Agentforce Testing Center
- The Agent API
- Canvas view in Agentforce Builder
Show answer & explanation
Correct: B — Agentforce Testing Center. The Testing, Deployment and Maintenance objective asks candidates to test an agent using Testing Center and to explain how its evaluations work. Systematic, repeatable evaluation before promotion is precisely its purpose.
Why not the others: Prompt Builder preview (A) checks a single template's output, not agent behaviour across scenarios. The Agent API (C) is an interface for interacting with agents programmatically. Canvas view (D) is an authoring view.
Source: Salesforce — Exam guide, Testing Center and deployment →An agent has been live for a month and stakeholders want to know which conversations are being resolved successfully and where the agent struggles. Which area of the current blueprint covers this?
- Multi-Agent Orchestration, through A2A telemetry
- Prompt Engineering, through template activation history
- Data 360 Fundamentals, through retriever tuning
- Governance and Observability, through agent analytics and optimization
Show answer & explanation
Correct: D — Governance and Observability. This domain covers the process for managing and monitoring agents plus agent analytics and agent optimization. In the current outline it is a distinct 10% objective rather than part of a combined lifecycle domain, which is a change from the earlier structure.
Why not the others: Multi-Agent Orchestration (A) concerns architecture choices and protocols such as MCP and A2A. Prompt Engineering (B) covers template design and governance, not production monitoring. Data 360 (C) covers grounding rather than agent performance measurement.
Source: Salesforce — Exam guide, Governance and Observability →An architect is asked about open standards that let agents interoperate with external tools and with other agents. Which two protocols does the current outline name?
- SOAP and REST
- OAuth and SAML
- Model Context Protocol and Agent-to-Agent communication
- GraphQL and gRPC
Show answer & explanation
Correct: C — MCP and A2A. The Multi-Agent Orchestration objective asks candidates to explain the purpose of existing open standard multi-agent protocols such as Model Context Protocol and Agent-to-Agent communication, alongside judging when a Single-Agent (SOMA) architecture is appropriate. This domain is new to the current outline.
Why not the others: SOAP and REST (A), OAuth and SAML (B), and GraphQL and gRPC (D) are all general integration or authentication technologies. None is a multi-agent orchestration protocol, and none is named in this objective.
Source: Salesforce — Exam guide, Multi-Agent Orchestration → Further reading: PowerKram — Certified Platform Developer exam →Keep going: Learning & Career resources
Agentforce sits at the centre of Salesforce's current direction, and the credential rewards breadth across configuration, data, and governance. Two PowerKram hubs back this exam.
Deep dive: exam format, what changed in the current outline, study path, and maintenance
Exam format and how it is scored
Sixty multiple-choice questions in 105 minutes, plus up to five unscored questions randomly integrated to gather performance data — those have no impact on your result. The passing score is 72%, which is high by Salesforce standards and leaves room for roughly sixteen wrong answers out of sixty. Registration is USD 200 and the retake is USD 100, plus applicable taxes. The exam is offered in English, French, and Japanese, and no reference materials may be used. Read the official exam guide →
What changed in the current outline
Four changes matter if you prepared earlier. The exam moved from five domains to six. Data Cloud for Agentforce became Data 360 Fundamentals, following the product rename. The former Development Lifecycle domain was split into Testing, Deployment and Maintenance at 10% and Governance and Observability at 10%, which separates pre-production work from production monitoring. And Multi-Agent Orchestration arrived as a new 5% domain covering MCP, A2A, and the Single-Agent (SOMA) architecture decision. Alongside those, the authoring model shifted: Agentforce Builder with Agent Script is now the primary path, and hybrid reasoning is explicitly examinable. Salesforce Architect: hybrid reasoning explained →
Why Agent Script matters more than its word count suggests
Agent Script is a compiled, declarative language. You do not write code that runs line by line; the script compiles into an Agent Graph, a structured specification consumed by the Atlas Reasoning Engine. That architecture is what makes reflective loops, multi-topic flows, and multi-agent coordination behave exactly as defined rather than being re-decided by the model on every turn. Understanding that compilation step is the difference between answering scenario questions confidently and guessing, because it explains why deterministic mechanisms such as filters, variables, and template expressions work the way they do. Introducing hybrid reasoning with Agent Script →
Where candidates lose marks
Three patterns recur. First, studying an outdated blueprint — the five-domain structure is still widely published, and anyone using it walks in with no preparation at all for Multi-Agent Orchestration and a blurred view of the lifecycle domains. Second, over-preparing on machine learning theory: the guide explicitly excludes LLM fine-tuning, transformer architecture, and external AI tooling, so time spent there is wasted. Third, underestimating the security context question type — several AI Agents items hinge on which user an agent runs as and what that means for action execution, which is configuration knowledge rather than AI knowledge.
Realistic study path
Most candidates with platform experience need four to eight weeks. A workable sequence: read the current exam outline and note the six domains explicitly, since that alone corrects most stale material; earn Agentblazer status on Trailhead, which Salesforce recommends directly; then build an agent end to end in a Developer org — author it in Agentforce Builder, switch between Canvas and Script view until Agent Script reads naturally, ground it in a Data Library with chunked unstructured content, evaluate it in Testing Center, and deploy it to another org. The two 10% lifecycle domains reward having actually done this once. Earn Agentblazer status on Trailhead →
Maintenance: this credential expires if you ignore it
You must complete the Agentforce Specialist maintenance module on Trailhead once a year, and the certification expires if you miss the due date. Given how quickly Agentforce changes across releases, the maintenance module doubles as the cheapest way to stay current — it is short and free, and it tracks the release changes you would otherwise have to chase yourself.
Where this certification leads
Agentforce Specialist is the credential Salesforce is building its catalogue around. While 24 other certifications are being retired and 16 renamed — most of the renames adding “Agentforce” to the title — this exam sits squarely in the direction of travel alongside Data 360 and the core platform. It pairs naturally with Certified Data Cloud Consultant for the grounding and retrieval side, and with Certified Platform Administrator as the configuration foundation Salesforce itself recommends.
Frequently asked questions about the Agentforce Specialist exam
What are the Agentforce Specialist exam domains and weights?
Has the Agentforce Specialist blueprint changed recently?
What is the passing score for the AI-201 exam?
Do I need to know Apex or machine learning theory?
What is Agent Script and do I need it for the exam?
Does the Agentforce Specialist credential expire?
Start your free 24-hour Agentforce Specialist practice trial
Full access to the question bank, both study modes, source-linked explanations, and score-by-domain reporting against the current six-domain outline. No credit card required.
Start free trial →