Salesforce Interview Questions with Answers Part 78

HomeMagulan Duraipandian

Salesforce Interview Questions with Answers Part 78

Magulan Duraipandian

1. How do you enforce both object- and field-level security in Salesforce Apex?

Answer: Use the stripInaccessible method to enforce both object- and field-level security. It automatically removes any fields or relationship fields from query/sub-query results that the current user isn’t allowed to see, and it cleans up sObjects—whether they’re about to be inserted/updated or were just deserialized from an untrusted source—so inaccessible fields never reach DML and cause exceptions.

Reference: https://developer.salesforce.com/docs/atlas.en-us.apexcode.meta/apexcode/apex_classes_with_security_stripInaccessible.htm

2. How do you control serialization using @JsonAccess in Apex?

Answer: You can restrict how your Apex class is serialized or deserialized by adding the @JsonAccess annotation to the class definition. This is useful for protecting sensitive data handling in JSON or XML operations.

You must configure the serializable and/or deserializable arguments using one of these four scopes:

  • always
  • never
  • sameNamespace
  • samePackage (Second-generation packages only)

Note: Failing to meet these access requirements at runtime will cause a JSONException.

Reference: https://developer.salesforce.com/docs/atlas.en-us.apexcode.meta/apexcode/apex_classes_annotation_JsonAccess.htm

3. How does the Atlas Reasoning Engine decide which action handler to execute when multiple actions are associated with a single subagent?

Answer: When a single subagent (or topic) in Salesforce’s Agentforce has multiple action handlers associated with it, the Atlas Reasoning Engine relies on a deliberative System-2 reasoning loop (ReAct framework).

4. What is the step-by-step process Salesforce Atlas uses to select an action handler?

Answer:

Plaintext

[ User Request ] 
       │
       ▼
1. Context Assembly (User message + Conversation History + Topic Scope)
       │
       ▼
2. Action Schema Evaluation (Semantic Intent Matching vs. Action Descriptions)
       │
       ▼
3. Rule & Guardrail Verification (Instructions + Preconditions + Permissions)
       │
       ▼
4. Action Execution (Invokes specific Apex / Flow / API / Prompt Template)
       │
       ▼
5. Observation Loop (Evaluates execution output before deciding next action)

5. How do you use Salesforce Data Cloud to provide real-time grounding and context to an AI agent?

Answer: You can leverage Data Cloud to rapidly retrieve similar contextual data (such as past cases) and ground the AI agent with real-time enterprise context.

Reference: https://www.infallibletechie.com/2025/10/find-similar-salesforce-cases-fast-with-data-cloud.html

6. What is the operational difference between configuring an Agent Action versus triggering a Salesforce Flow?

Answer: Use Flow for tasks that require predictability and compliance. Use Agent Actions when you need contextual understanding and multi-step reasoning.

Use Salesforce Flow directly when:

  • Executing strict compliance tasks (e.g., financial ledger entries, automated discount approvals).
  • Handling backend system integrations where data inputs are always structured and predictable.
  • Running scheduled batch jobs or record-triggered field updates.

Use an Agent Action when:

  • Exposing an existing process (like an Autolaunched Flow or Apex class) to an AI agent.
  • Interacting with humans through open-ended conversations where inputs arrive in unstructured text.
  • Enabling dynamic, multi-step actions where the execution sequence depends on what information the user provides turn-by-turn.

7. How do you design, test, and manage Agentforce Assets to ensure consistency across multiple sandboxes?

Answer: Salesforce Agentforce Testing Center can be used. Use Salesforce’s native Agentforce Testing Center to create automated test suites. Define sets of sample user prompts and set expected topic/action execution targets. Run these test suites in each target sandbox immediately after metadata deployment to detect regression.

8. What is the Salesforce Einstein Trust Layer and what key features does it provide?

Answer: The Einstein Trust Layer is a secure AI architecture built into the Salesforce platform designed to let you safely use generative AI and large language models (LLMs). It acts as a protective shield between your company’s CRM data and external AI models, preventing data leaks, toxicity, and model hallucinations.

Key capabilities include:

  • Zero Data Retention
  • Grounding
  • Data Masking
  • Audit Trails: Tracks and logs every prompt, response, and toxicity score in the Data Cloud, providing transparency and accountability for AI usage.

9. What exact guardrails and audit trails do you implement to govern Salesforce Agentforce behavior and prevent AI hallucinations?

Answer:

  1. Clear Agent Instructions
  2. Brief description for Variables and Agent Actions
  3. RAG
  4. Sample Intents (examples)
  5. before_reasoning and after_reasoning blocks in Agent Script

Reference: https://www.infallibletechie.com/2026/04/salesforce-agentforce-running-pre-reasoning-actions-in-agent-script.html

10. How would you architect a hybrid routing model using Omni-Channel to seamlessly pass a customer from an AI agent to a human agent?

Answer: Use a Custom Escalation topic.

References:

11. When designing a high-volume integration, how do you decide whether to process data natively in Salesforce or offload it to AWS microservices?

Answer:

  • NATIVE SALESFORCE (Data Cloud / Apex): Is it for Human/CRM Workflows? (UI visibility, operational triggers)
  • AWS MICROSERVICES (Lambda / EventBridge / ECS / S3 / SQS): Is it Heavy Computing / Analytics? (Data transformation, logs, payload parsing)

12. How does variable visibility work in Salesforce Agentforce Agent Script?

Answer: The visibility property for the variable is optional. The default value is Internal. You can set visibility to External to allow an API to set the variable’s value, or to change the variable’s value when testing the agent in simulate mode.

13. What is Salesforce Agentforce Agent Memory and how do you enable it?

Answer: Salesforce Agentforce Agent Memory lets an agent remember relevant details from past conversations and reuse them as context in future ones. Instead of asking users to repeat themselves, the agent gives more relevant, personalized responses.

For 50 past conversations, use:

YAML

context:
  memory:
    enabled: True

For user profile memory, use:

YAML

context:
  user_profile:
    enabled: True

(Reference: salesforce-agentforce-agent-memory.html)

14. What is the use of reset_to_initial_node in Salesforce Agentforce Agent Script?

Answer: It restarts each customer turn at the start_agent block instead of resuming from where the previous turn left off.

When to Use:

  • Enable: Stateless, single-turn Q&A or planner agents that need to re-evaluate from scratch every turn.
  • Disable (Default): Multi-turn conversational flows that need to retain state and resume mid-conversation.

15. What is the relationship between Headless 360 and AIforce in Salesforce?

Answer: At Dreamforce 2026, Headless 360 was rebranded to AIforce.

Leave a Reply