Home / Agentforce Specialist

Salesforce · Agentforce Specialist

Agentforce Specialist Certification Practice Questions

Free Agentforce Specialist certification questions from all 6 official domains, with every answer option explained.

Start practicing free Take a mock examSee plans

Salesforce Certified Agentforce Specialist

Questions
60

Multiple-choice.

Time
105 min
To pass
72%
Exam Price
$200

Retake $100, plus applicable taxes.

What is on the Agentforce Specialist exam?

Open a domain for its objectives and their practice questions.

Try free Agentforce Specialist practice questions

Pick an answer to see why each option is right or wrong. A preview of 4 of the 12 free questions, which cover every domain.

Domain 1: Prompt Engineering

1.1 Given business requirements, identify when it's appropriate to use Prompt Buildereasy

A month after a prompt template goes live, an admin wants to track its token usage and response time and compare results between two of its versions. Where can the admin do this?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Prompt Builder covers the life of a template after deployment too: its built-in tools track token usage and response time and compare versions or models.

The Start panel sets the flow's inputs and the template type it serves. It has no monitoring of how the template performs.

Under zero data retention, external providers keep no prompts or responses, so there is no provider-side history to inspect.

Lightning App Builder is where a Field Generation template is assigned to a field; it does not report token usage or response times.

1.2 Identify access controls governing prompt templateseasy

A rep generates an AI reply from a prompt template that references account fields the rep's profile cannot read. How does secure data retrieval handle those fields during dynamic grounding?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Running as an admin would leak data past the rep's permissions, which is what secure data retrieval exists to prevent.

The Trust Layer does not change permissions. The retrieved data contains nothing that needs escalated permissions.

Grounding runs in the user's context: secure data retrieval honors existing object and field permissions, so the prompt only receives data the rep is authorized to see.

Masking hides sensitive values such as names from the LLM; it is not a permission control. Data the rep can't access is never retrieved in the first place.

Domain 2: Data 360 Fundamentals

2.2 Explain foundational concepts of Data 360 such as chunking, indexing, and retrieverseasy

A prompt template needs content from three separate search indexes in a single call. Which kind of retriever searches all three in parallel?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Enrichment adds metadata and question chunks inside one index; it doesn't combine separate indexes.

Hybrid describes a search index that mixes vector and keyword search over one index, not a retriever that spans several indexes.

An ensemble retriever bundles individual retrievers and runs their searches across multiple sources at the same time.

A data library's retriever serves that one library's single data source, so it can't span three indexes.

2.1 Explain the considerations of Agentforce Data Library and its conceptseasy

At runtime, a user's question has just been added to a prompt template that references a data library's retriever. What happens next?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

With the query in the template, the retriever's link is used to search the index, and the best matches are added back into the prompt before the LLM is called.

Chunking is a real step, but it happens once during setup, before the index and retriever exist. At runtime the content is already chunked and indexed.

Data streams bring source content into Data 360 at setup. User questions aren't ingested into the index this way.

The retriever supplies data before generation; it doesn't review the LLM's output afterwards.

The other 8 free questions, by domain:

Domain 3: AI Agents · 2 questions
3.2 Explain the components and benefits of hybrid reasoning, including how Agent Script functions in Canvas and Script Vieweasy

A developer edits a subagent directly in Script view and saves. Now the builder won't switch back to Canvas view, and the console lists an undefined reference. What should the developer do?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Some Script view edits can make Canvas unavailable; clearing the errors in Script view is what restores the Canvas view.

Preview modes test the agent's behavior; they don't repair the script or change which editing view is available.

Canvas is the view that's unavailable, so it can't be used to work around the problem until the script errors are fixed.

Committing saves a snapshot, so it sounds like a way to reset the view, but an agent with errors can't be committed at all.

3.1 Explain how an agent works and its basic building blocks of agent scripteasy

Version 2 of an agent is live. An admin commits version 3 and activates it. What happens to version 2?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

To change a committed version you save it as a new draft; activating another version doesn't convert the old one back into a draft.

No deletion is needed; activation itself switches the old version off, and committed versions remain available as snapshots.

An agent can have only one live version, so activating the newly committed version automatically deactivates the one that was live before.

Versions don't split traffic; the lifecycle allows a single live version at a time, so the previous one stops serving conversations.

Domain 4: Testing, Deployment, and Maintenance · 2 questions
4.1 Given a scenario, test an agent using Testing Centereasy

A team is ready to batch test a new service agent that creates and updates case records. In which environment should they run their Testing Center suites, and why?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Realistic data sounds attractive, but it's exactly why production is the wrong place: the agent's actions would modify live customer records.

Cost is a real concern, but tests consume requests and credits the same way whether they run in a sandbox or not, so this gives no reason to test in production.

Agent tests execute real actions, so they can create or modify data. Running them in a sandbox keeps those changes away from production records.

Sandboxes now have access to Data 360 and Agentforce for agentic testing, so this premise is false.

4.2 Explain how Testing Center evaluations workeasy

In a Testing Center results row, what does the Subagent Evaluation column tell you?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

That comparison is Response Evaluation, a separate column that judges the reply, not the routing.

That is Action Evaluation, which checks the actions run within the subagent rather than which subagent was chosen.

A 1 to 5 Quality Score comes from Agent Analytics in production monitoring; Testing Center evaluations report Pass or Fail.

Subagent Evaluation compares the actual subagent picked for the utterance with the Expected Subagent value and reports Pass or Fail.

Domain 5: Governance and Observability · 2 questions
5.2 Explain agent analytics and agent optimizationeasy

Which Agent Analytics metric rises when more customers leave a chat untouched until it times out?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

Escalation Rate counts sessions handed to a person or another agent; a timed-out chat wasn't handed over.

Success Rate looks at interactions with action steps that finished without errors, not at idle sessions.

Deflection counts sessions that ended without escalation at the user's request; more timeouts don't raise it.

Abandonment Rate is the share of sessions that timed out or ended without engagement, so idle chats that time out push it up.

5.1 Explain the process for managing and monitoring agentseasy

An agent action that calls an external inventory API fails intermittently with HTTP 429 during peak hours. Which fix matches this error?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

429 means the external system is rate-limiting the calls; retrying with backoff spreads requests out until the limit clears.

Refreshing the token is the fix for 401 Unauthorized, not for being rate-limited.

A wrong endpoint or record ID produces 404 Not Found, and it would fail every time, not just at peak hours.

Permission problems show up as 403 Forbidden, not as a rate limit.

Domain 6: Multi-Agent Orchestration · 2 questions
6.2 Explain the purpose of existing open standard multi-agent protocols such as MCP and A2Aeasy

An admin asks what problem the Model Context Protocol (MCP) solves for Agentforce agents. Which description is accurate?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

MCP does the opposite: its tools surface as agent actions so the agent can call more external functions.

This is a real open protocol, but it is A2A, which covers agent-to-agent discovery and collaboration. MCP is about an agent reaching tools and data, not other agents.

How agents are organized is an architecture question (SOMA or MOMA). MCP is a protocol, the language an agent uses to reach external tools.

MCP is an open, vendor-neutral standard for connecting agents to external capabilities. A server exposes tools, resources and prompts, and the Agentforce MCP client uses them without custom integration code.

6.1 Given a scenario, determine whether a Multi Agent architecture is appropriate for scalability and controleasy

A retail service team with all its agents in one Salesforce org is adopting Single-Org Multi-Agent (SOMA). What experience does this architecture give the end users?

Included in Sample: sample every domain.See plans

Choose an answer to check it. Afterward, select any option to review its explanation.

This is the situation SOMA is meant to remove: users managing conversations with several agents one by one. The orchestrator hides the specialists behind one touchpoint instead.

SOMA puts a parent orchestrator in front of the user and delegates each task to the specialist agent best suited to it, so the user keeps a single touchpoint while specialists work in the background inside the same trust boundary.

SOMA is built on delegation from a parent agent to specialists; keeping every topic in one agent is the single-agent model whose limited cognitive span SOMA is meant to work around.

Tempting because it describes a real multi-agent pattern, but delegating to agents in other orgs is MOMA. This team has every agent in one org, which is exactly the SOMA case.

Practice all 12 free questionsGet 150+ questions

Agentforce Specialist question bank coverage

DomainExam weightFree sampleMaster (full bank)
1. Prompt Engineering20%236
2. Data 360 Fundamentals20%236
3. AI Agents35%263
4. Testing, Deployment, and Maintenance10%218
5. Governance and Observability10%218
6. Multi-Agent Orchestration5%210
Total100%12181

Looking for a Agentforce Specialist practice exam? Start with the free questions above, then take the timed mock exam to rehearse the Agentforce Specialist format under exam conditions.

How these Agentforce Specialist questions are made

The domains, weights and objectives come from Salesforce's official Agentforce Specialist exam guide (vSpring '26). Questions are original and written only from the vendor's official prep material, and each one cites the passage it is based on. Every question is then answered independently, without seeing its key, and every answer option's explanation is checked for accuracy before it is published.