Agentforce Specialist Domains Explained

The short version

The exam has six sections. AI Agents is 35%, Prompt Engineering and Data 360 Fundamentals are 20% each, Testing/Deployment/Maintenance and Governance/Observability are 10% each, and Multi-Agent Orchestration is 5%. The topic lists come from the Salesforce exam guide (Spring ’26 alignment, checked October 18, 2026). The traps and practice tips are our own general advice.

AI Agents (35%)

What it covers. How agents work and the building blocks of Agent Script; hybrid reasoning, including Agent Script in Canvas and Script View; deterministic behavior using filters, variables, and template expressions; standard and custom topics and actions; channels such as digital experience, email, voice, and Slack; the security context an agent runs in; Employee versus Service agents; and when the Agent API is appropriate.

In practice. Hybrid reasoning means some steps are left to the language model and some are fixed. The deterministic tools let you force an outcome where the business cannot accept variation. Security context decides which records and actions an agent may touch.

Typical traps.

  • Picking an Employee agent for unauthenticated customers. Employee agents run for signed-in users; customer-facing agents are a different type.
  • Using the model for something that must be exact. If the scenario says “always” or “never”, look for a deterministic option.
  • Confusing channels. Match the channel to where the conversation happens.

Practice. Build one Employee agent and one Service agent. Add a standard and a custom action. Read a sample of Agent Script and say what each block does.

Prompt Engineering (20%)

What it covers. When Prompt Builder fits; access controls for templates; template types such as field generation and flex; choosing a grounding technique; creating, activating, and running templates; writing effective prompts; the Trust Layer’s security and privacy features; and controlling access to models.

Typical traps.

  • Using a retriever when the question needs one exact stored value. A record merge field delivers the field itself; a retriever returns passages judged relevant.
  • Forgetting that grounding methods can be combined in one template.
  • Treating the Trust Layer as a single feature. It is a set of protections, so learn what each one does.

Practice. Create a field-generation template and a flex template. Ground one with merge fields and another with a retriever, and compare outputs.

Data 360 Fundamentals (20%)

What it covers. The considerations and concepts of the Agentforce Data Library, and the basics of chunking, indexing, and retrievers.

In practice. Unstructured content is split into chunks, indexed for search, and returned by a retriever. The library packages these steps for you.

Typical traps.

  • Assuming the library works without Data 360 set up first.
  • Mixing up what the library does for you and what stays manual, such as access and channel configuration.
  • Confusing a search index with a retriever. One stores searchable data, the other fetches from it.

Practice. Create a library from a few documents and inspect what it generates.

Testing, Deployment, and Maintenance (10%)

What it covers. Testing an agent with Testing Center, how its evaluations work, and what to consider when deploying an agent or a template from a sandbox to production.

Typical traps. Not knowing which part of a test failed, and forgetting that a template may be a dependency of the agent you deploy. Read each scenario for what moves and what must already exist in the target org.

Practice. Build a small test suite, break one expected result on purpose, and read the report.

Governance and Observability (10%)

What it covers. Managing and monitoring agents, plus agent analytics and optimization.

Typical traps. Mixing up tools that look similar. Ask whether the scenario needs real-time alerts, long-term performance reporting, or an explanation of a single decision. Also note that observability features can depend on other platform pieces being provisioned first.

Practice. List each monitoring tool with one sentence on the question it answers.

Multi-Agent Orchestration (5%)

What it covers. Deciding whether a multi-agent architecture suits a scenario’s scale and control needs, and the purpose of open protocols such as MCP and A2A.

Typical traps. Reaching for multi-agent when one well-organized agent is enough. It is the right answer when growth makes a single agent hard to keep reliable. Know the purpose of each protocol in one sentence.

Practice. Draw a primary agent delegating to two specialists, and say what each protocol connects.

More guides for AGENTFORCE-SPECIALIST

Next step

Test your recall with the sample questions. For a section-by-section score, try our Agentforce Specialist practice tests.

Frequently asked questions

Which section has the most questions?

AI Agents, at 35% of the exam.

Is Multi-Agent Orchestration worth studying at 5%?

Yes, briefly. It is a small, learnable section, and the protocol vocabulary also helps elsewhere.

Do the weightings change?

They can with new outlines. Check the exam guide before you book.

Try these practice exams