Salesforce Certified Agentforce Specialist (AI-201) Practice Exam
Prepare for the current Agentforce Specialist exam across all five domains — AI Agents, Prompt Engineering, Data Cloud for Agentforce, Development Lifecycle, and Multi-Agent Interoperability — with scenario questions, source-linked explanations from Salesforce docs, and full timed simulation in Exam mode.
Start 24-hour free trial →Agentforce Specialist exam at a glance
- Vendor
- Salesforce
- Exam code
- AI-201 (current version)
- Certification
- Salesforce Certified Agentforce Specialist
- Level
- Specialist (manage and optimize Agentforce)
- Blueprint
- Agentforce Specialist Exam Guide (revised October 2025 five-domain structure)
- Format
- Multiple choice / multiple select; scenario-based
- Questions
- 60 scored, plus up to 5 unscored (65 total)
- Duration
- 105 minutes
- Passing score
- 73%
- Delivery
- Onsite test center or online proctored
- Cost (USD)
- $200; retake $100 (regional pricing varies)
- Prerequisites
- None. Salesforce recommends hands-on Agentforce experience; the Administrator or Platform App Builder credential helps.
- Languages
- English, Japanese
Sources: Salesforce — Agentforce Specialist credential page and the official Agentforce Specialist Exam Guide. Verify current exam details with Salesforce before scheduling; Agentforce evolves each release.
About the Salesforce Certified Agentforce Specialist certification
The Agentforce Specialist credential is Salesforce’s core “you can build and manage production AI agents” certification. It validates that you can configure Agentforce end to end — designing agents in Agent Builder, writing effective prompts in Prompt Builder, grounding responses in enterprise data through Data 360 (formerly Data Cloud), securing agents with the Agent User model, testing in the Agentforce Testing Center, and deploying and monitoring agents across channels.
Two things are essential to know before you study. First, the current exam code is AI-201, and the blueprint was significantly revised in October 2025: the older Service Cloud and Sales Cloud sections were retired and folded into a larger AI Agents domain, Agent Lifecycle became its own domain, and a new Multi-Agent Interoperability domain (MCP, Agent-to-Agent, and Agent API) was added. Second, this is a scenario-driven exam — questions describe a business situation and ask for the best configuration, with distractors that are wrong for specific, learnable reasons. Rote definitions are not enough; hands-on Agentforce practice is.
PowerKram’s Agentforce Specialist practice questions mirror that scenario format and link each explanation to the exact Salesforce documentation page it derives from, so a wrong answer turns into a specific page to read. For the broader Salesforce ecosystem context this credential sits in, see our Salesforce definitive guide.
Agentforce Specialist (AI-201) exam domains and weights
The revised AI-201 blueprint splits the exam into the five weighted domains below. AI Agents is by far the heaviest — it absorbed the retired Service and Sales Cloud material — so it is where most candidates should invest first. Plan your study roughly in proportion to these weights.
Agent Builder configuration; standard and custom subagents (formerly topics) and actions via Flow, Apex, and prompt templates; the Atlas Reasoning Engine; agent types (Employee, Service, Sales); the Agent User security model and permission sets; connecting agents to channels such as Slack, email, and digital experiences; deterministic behavior with filters and variables.
When to use Prompt Builder; user roles to manage and execute templates; field-generation and flex prompt types; grounding techniques (merge fields, related lists, retrieval); creating, activating, and executing templates; best practices for clear, effective prompts that reduce hallucinations.
Agentforce Data Library types; grounding agents in structured and unstructured data; chunking and indexing unstructured content; retrievers (individual and ensemble); choosing a search type (keyword, vector, or hybrid) and how it affects response accuracy. (Data Cloud is now Data 360.)
Building, testing, deploying, and monitoring agents across environments; validating behavior in the Agentforce Testing Center; deploying safely from sandbox to production; tracking adoption and iterating on instructions to improve accuracy over time.
Model Context Protocol (MCP) for connecting agents to external systems and data; Agent-to-Agent (A2A) protocol for agents exchanging context and delegating tasks; the Agent API for programmatic access from external systems. The newest domain, and the smallest by weight.
Weights reflect the current AI-201 blueprint (AI Agents 35%, Prompt Engineering 20%, Data Cloud for Agentforce 20%, Development Lifecycle 20%, Multi-Agent Interoperability 5%). Confirm the live weighting on the official Agentforce Specialist credential page before your exam, as Salesforce updates the blueprint with major releases.
Who the Agentforce Specialist exam is for
Salesforce positions this credential for people who manage and optimize Agentforce — a blend of admin, builder, and AI-configuration skills. You do not need to be a coder or an AI engineer, but hands-on Agentforce experience is essential:
- Salesforce admins and app builders extending their skills into AI, taking the lead on their organization’s Agentforce implementation.
- Consultants and implementers deploying autonomous agents for clients across Service, Sales, and employee use cases.
- Business analysts and solution designers who scope agent use cases, define success metrics, and translate business needs into agent configuration.
- Developers who pair prompt engineering and custom Apex/Flow actions with the declarative Agent Builder to solve harder scenarios.
If you are brand new to Salesforce, the Platform Administrator credential is a sensible foundation first. Because grounding agents in enterprise data is a whole exam domain, the Data Cloud (Data 360) Consultant exam is a strong companion. For the roles this credential supports — with salary ranges and progression — see the AI engineer career path in our Career Hub.
What this Agentforce Specialist practice exam delivers
Learn mode
Get the correct answer, the reasoning, and why the other options fail — immediately after each question. Best for the AI Agents domain, where subagent/action design and the Atlas Reasoning Engine drive the “best configuration” call.
Exam mode
60 questions, 105-minute timer — the real AI-201 format, including multi-select items. Build the pacing and scenario-reading speed the timed exam demands.
Source-linked explanations
Every answer cites the exact Salesforce documentation page it was built from — so you learn from Salesforce’s own docs on Agent Builder, Prompt Builder, Data 360, and MCP, not just a memorized letter.
Score by domain
Results break down by the five current domains — AI Agents, Prompt Engineering, Data Cloud for Agentforce, Development Lifecycle, and Multi-Agent Interoperability — so practice tells you exactly where to focus.
Sample Agentforce Specialist practice questions
Ten free scenario questions across the five current AI-201 domains, each with a full explanation and a source link to the Salesforce documentation it derives from. The complete bank is available with the 24-hour trial.
A service team handles 5,000 cases a month; about 60% are routine order-status, returns, and account-update questions. The director wants an Agentforce agent to take the routine load. How should the specialist prioritize the use cases?
- Analyze case data to find the highest-volume, lowest-complexity categories, confirm they can be answered from existing data sources, build the agent to handle those first with clear escalation to humans for complex cases, and define deflection-rate and CSAT success metrics
- Deploy the agent to handle all 5,000 monthly cases, including complex technical issues, from day one
- Start with the most complex use cases to demonstrate maximum AI value
- Wait until the agent can handle 100% of cases before deploying anything
Show answer & explanation
Correct: A — High-volume, low-complexity first, with escalation and metrics. Starting with routine, data-answerable categories gives the best ROI at the lowest risk, escalation keeps complex cases with humans, and defined metrics let you prove value and iterate.
Why not the others: Handling all cases at once (B) risks poor responses on complex issues. Complex-first (C) is high-risk with little volume benefit. Waiting for perfection (D) delays all value indefinitely.
Source: Salesforce Architect — Agentic patterns → Further reading: PowerKram — Salesforce Definitive Guide →A specialist is building an agent in Agent Builder that must look up order status by order number or customer email. How should the order-lookup action be configured?
- Give the agent direct database access to write SOQL in real time
- Create a Flow-based action that accepts order number or email, queries the Order object, and returns formatted status, shipping, and delivery details, then instruct the subagent to invoke it on order-status intent
- Pre-load all order data into the agent’s prompt context
- Have the agent tell customers to check the website for order status
Show answer & explanation
Correct: B — A Flow-based action invoked by intent. Flow actions give controlled, secure data access with defined inputs and outputs that the agent calls when it recognizes the matching intent — the intended pattern for structured lookups in Agent Builder.
Why not the others: Direct database access (A) bypasses security controls. Pre-loading all orders (C) is impractical and blows past context limits. Redirecting to the website (D) defeats the agent’s purpose.
Source: Salesforce Help — Agentforce Builder → Further reading: PowerKram — Salesforce Definitive Guide →A specialist must build a prompt template that generates personalized replies to customer complaints, referencing each customer’s case details and current company policy. How should the template be designed?
- In Prompt Builder, use merge fields to insert case details (case number, issue type, name) from record context, ground the prompt on current policy knowledge articles, add tone guidelines, and test across diverse case scenarios
- Write a single static prompt with no dynamic data or grounding
- Let the model generate replies freely with no prompt engineering
- Create a separate prompt template for every possible complaint type
Show answer & explanation
Correct: A — Merge fields plus grounding on current policy. Dynamic merge fields personalize each reply, grounding on knowledge articles keeps policy accurate, and tone guidelines hold brand voice — then testing across scenarios validates it.
Why not the others: A static prompt (B) yields generic replies. Unguided generation (C) risks inaccurate, off-brand output. Per-complaint templates (D) are unmaintainable as complaint types grow.
Source: Salesforce Help — Prompt templates → Further reading: PowerKram — Salesforce Definitive Guide →After deployment, an agent is giving customers outdated return-policy information — the policy changed last month. How should the specialist fix the inaccurate responses?
- Disable the agent entirely until policies stop changing
- Add a disclaimer to every response saying information may be outdated
- Trace the grounding to find where the outdated policy is retrieved, update the knowledge articles / Data Library source with current policy, confirm the retriever references the correct source, and re-test in Conversation Preview
- Retrain the entire underlying AI model with the new policy
Show answer & explanation
Correct: C — Fix the grounded data source, then re-test. Agentforce agents ground responses in your data (via the Data Library / Data 360), so updating the stale source and confirming the retriever points at it directly fixes accuracy; Conversation Preview validates the fix.
Why not the others: Disabling the agent (A) removes all value. Disclaimers (B) erode trust without fixing anything. Agentforce uses grounded data, not a custom-trained model, so retraining (D) is the wrong mental model.
Source: Salesforce Architect — Grounding with Data 360 → Further reading: PowerKram — Salesforce Definitive Guide →An agent must read Case details and update Case status, but must never modify Account financial information. How should the specialist configure the agent’s access?
- Give the Agent User a dedicated permission set granting Read on Account and Case, Edit on the Case Status field only, and no access to Account financial fields — following least privilege
- Give the agent System Administrator permissions for maximum flexibility
- Configure no special permissions and assume the agent inherits the running user’s access
- Grant full Read/Write on all objects to prevent any permission errors
Show answer & explanation
Correct: A — A least-privilege permission set on the Agent User. The Agent User’s access is set explicitly through permission sets; granting only what the task needs (read Case/Account, edit Case Status, no financial fields) enforces least privilege.
Why not the others: System Admin (B) grants far too much. Assuming inheritance (C) is wrong — agent access must be explicitly configured. Full Read/Write (D) violates least privilege and the stated restriction.
Source: Salesforce Help — Agent User permissions best practices →Before deploying an agent to production, a specialist needs to validate that it responds correctly across a broad, repeatable set of scenarios rather than testing ad hoc. Which Agentforce capability fits?
- Build test cases in the Agentforce Testing Center to run the agent against many scenarios, review pass/fail results, and iterate before deploying from sandbox to production
- Ask a few colleagues to chat with the agent once and call it done
- Deploy straight to production and fix issues as customers report them
- Only spot-check the single happy-path conversation
Show answer & explanation
Correct: A — The Agentforce Testing Center. Testing Center runs the agent against a defined set of scenarios with repeatable pass/fail evaluation, which is the intended way to validate reliability before a sandbox-to-production deployment.
Why not the others: A few one-off chats (B) and happy-path spot checks (D) miss edge cases. Deploying first and fixing later (C) exposes customers to defects and undermines trust.
Source: Salesforce Help — Agentforce Testing Center →A company wants one agent available on its website, mobile app, and Slack, with a consistent experience across all three. How should the specialist configure multi-channel deployment?
- Deploy one agent across channel-specific surfaces (embedded service for web, mobile, Slack), keep the same agent configuration and knowledge sources for consistency, and account for channel constraints like Slack message limits
- Build three separate agents, one per channel
- Deploy only on the website and link users on other channels to it
- Wait until the agent can handle 100% of cases before deploying to any channel
Show answer & explanation
Correct: A — One agent, multiple channel surfaces. A single agent configuration deployed across channels keeps responses consistent, while channel-specific surfaces handle UX differences — the supported way to connect an agent to Slack, email, and web.
Why not the others: Three separate agents (B) create maintenance drift and inconsistency. Website-only (C) abandons users elsewhere. Waiting for 100% (D) blocks all rollout.
Source: Salesforce Help — Agent channel considerations →Thirty days after launch, the director wants to know whether the agent is actually reducing case volume while keeping customers satisfied. Which metrics should the specialist monitor and report?
- Only the total number of conversations the agent handled
- Deflection rate (resolved without human handoff), escalation rate, post-interaction CSAT, average handling time vs. human resolution, top unresolved query patterns, and conversation completion rate
- Success measured solely by the reduction in support headcount
- Wait six months before measuring anything
Show answer & explanation
Correct: B — A rounded set of adoption and quality metrics. Deflection and escalation rates show load impact, CSAT and handling-time show experience, and unresolved-pattern analysis shows where to improve the agent next — together they answer the director’s question and drive iteration.
Why not the others: Raw conversation count (A) says nothing about outcomes. Headcount reduction alone (C) is a narrow, risky proxy. Waiting six months (D) forgoes early course-correction.
Source: Salesforce Architect — Agent monitoring patterns →Customers say the agent’s answers are technically correct but sound robotic and impersonal. How should the specialist improve conversational quality?
- Refine the agent instructions with specific tone and personality guidelines and example exchanges, add guidance to acknowledge customer emotion before solving, and iterate using conversation logs from low-satisfaction interactions
- Remove all personality guidelines so the agent sounds more neutral
- Add lots of emojis and casual slang to seem more human
- Have the agent pretend to be a human and never disclose it is AI
Show answer & explanation
Correct: A — Tone guidelines, examples, empathy, and log-driven iteration. Clear tone instructions with example exchanges and empathy cues improve conversational quality, and reviewing low-satisfaction logs tells you what to refine next.
Why not the others: A neutral tone (B) worsens the robotic feel. Excessive emojis/slang (C) read as unprofessional. Concealing AI identity (D) violates transparency and erodes trust.
Source: Salesforce Help — Agentforce Builder (instructions) →A specialist must let an external system trigger and exchange data with an Agentforce agent programmatically — for example, so a company website can send a request to the agent and receive a structured response. Which approach fits?
- Hard-code the agent’s credentials into the external website’s front-end JavaScript
- Email a human to copy answers from the agent into the external system by hand
- Use the Agent API so external systems can invoke the agent and exchange structured requests and responses programmatically over REST
- Export a static list of answers nightly for the website to display
Show answer & explanation
Correct: C — The Agent API. The Agent API gives external systems secure, programmatic REST access to an Agentforce agent — the multi-agent-interoperability mechanism for letting outside applications invoke agents and receive structured responses.
Why not the others: Front-end credentials (A) are a serious security risk. A human copy-paste loop (B) is neither real-time nor scalable. Nightly static exports (D) give stale data, not live agent responses.
Source: Salesforce Developers — Agentforce / Agent API → Further reading: PowerKram — Salesforce Definitive Guide →Keep going: Learning & Career resources
Agentforce is one of the fastest-moving corners of the Salesforce ecosystem, and AI-agent skills are in high demand. Two PowerKram hubs back this exam.
Deep dive: AI-201 format, scoring, what changed in the 2025 revision, and study path
Exam format and scoring
The Agentforce Specialist exam (AI-201) has 60 scored questions plus up to 5 unscored pilot questions, delivered in 105 minutes. Questions are single- and multi-select and scenario-based; for multi-select items you must choose all correct options to earn the point, with no partial credit. The passing score is 73%, and you can take the exam onsite or online with a proctor. Because the exam rewards applying features to situations, hands-on practice in a Developer Edition org matters more than memorization. Read how Salesforce exams work →
What changed in the October 2025 revision
The blueprint was substantially rewritten. The AI Agents domain grew to 35% and absorbed the retired standalone Service Cloud and Sales Cloud sections; Prompt Engineering dropped from 30% to 20% and added prompt-quality topics; Data Cloud for Agentforce was rewritten around Data Libraries, chunking/indexing, retrievers, and search-type selection; Development Lifecycle became its own 20% domain; and a new Multi-Agent Interoperability domain (5%) introduced MCP, Agent-to-Agent, and the Agent API. If you study from pre-October-2025 material, you will be misaligned. See the Data Cloud (Data 360) Consultant exam →
Terminology to keep current
Two renames trip up candidates on older material. Data Cloud is now Data 360 (functionality unchanged). And beginning in 2026, agent topics are now called subagents — you may see both terms in transitional documentation. Core concepts to master by name include the Atlas Reasoning Engine (which plans and executes), the Einstein Trust Layer, and the Agent User security model. Read the Salesforce definitive guide →
Realistic study path
A common plan runs four to eight weeks: read Agentforce end to end for vocabulary fluency, then get hands-on in a Developer Edition org — build a basic Q&A agent and a more complex agent that mixes Apex and Flow actions; work through the AI Agents and Prompt Engineering material first since together they are more than half the exam; add Data 360 grounding, retrievers, and search types; then drill Development Lifecycle and the small Multi-Agent Interoperability domain. Finish with timed practice runs to confirm you clear 73% consistently. Start from the Platform Administrator exam →
Where this certification leads
Agentforce Specialist is one of the most in-demand new Salesforce credentials, and it stacks well: with the Administrator credential as a foundation, the Data 360 Consultant exam for grounding depth, and platform developer skills for custom actions. It supports roles building and managing enterprise AI agents — AI-focused Salesforce specialists, consultants, and solution designers. For salary ranges and progression, the Career Hub is the place to look. Career Hub — AI engineer roles →
Frequently asked questions about the Agentforce Specialist exam
What is the exam code and current blueprint for the Agentforce Specialist exam?
How many questions are on the exam and what is the passing score?
How much does the Agentforce Specialist exam cost?
Are there prerequisites for the Agentforce Specialist exam?
Is Einstein Copilot still on the exam?
Start your free 24-hour Agentforce Specialist practice trial
Full access to 500+ questions across all five current domains, both study modes, and source-linked explanations. No credit card required.
Start free trial →