Skip to main content

AI, Agentforce and Claude Interview Questions

215 practice questions from 61 vtfriday guides on AI fundamentals, Agentforce, and Claude with Salesforce. Each question links to the guide that answers it, with the answer just below the question.

How to practise: read a question and answer it out loud first, as you would in an interview. Then open the link and compare your answer with the guide's. Inside a guide, the Practice button hides every answer so you can test yourself, and remembers what you knew and what to review.

The questions are grouped by topic, then by guide, newest guides first.

AI fundamentals

How to Test AI Guardrails: The 6-Angle Attack Strategy

  1. Q: How do you genuinely know an AI guardrail actually works?

Agentforce

Salesforce Agentforce Pricing: Digital Wallet & Credit Consumption Explained

  1. Q: In an interview, if someone asks "What does it actually cost to run an Agentforce AI agent?", how should I answer?
  2. Q1: How do long reasoning chains directly drive up your Agentforce bill?
  3. Q2: What does an Agentforce "trace" (in the Plan Tracer / Event Logs) show you that a normal chat reply hides?
  4. Q3: Why is "Deflection Rate" by itself a dangerous metric for judging an AI agent?
  5. Q4: Where exactly do you go in Salesforce to check this credit consumption, and what permissions do you need?
  6. Q5: Does Digital Wallet update instantaneously the exact second I run a single test prompt?
  7. Q6: How can I set up automated alerts so we don't accidentally run out of Flex Credits mid-month?

Agentforce Studio Analytics vs. Optimization: How to Measure AI Success

  1. Q: In an interview, if someone asks "How do you know your Agentforce AI is still working properly once it's live in production?", how should I answer?
  2. Q1: If routing accuracy drops by 5%+ after deploying a new subagent or Topic, how exactly do you fix the "utterance overlap"?
  3. Q2: What is a "Knowledge Gap" inside Agentforce Optimization, and how do we fix it?
  4. Q3: What is the difference between a bad "Escalation Rate" and healthy "Escalation Quality"?
  5. Q4: Where is all this Agentforce Analytics and Optimization session data actually stored under the hood?
  6. Q5: Why should I read 10 raw failing conversation traces instead of just trusting the high-level summary charts?

Mastering Salesforce Agentforce Variables: Explicit Memory for AI Agents

  1. Q: How does an agent remember something across a long conversation?
  2. Q: Where is the best place to set initial variable values?
  3. Q: What is the functional difference between mutable and linked variables?

🚀 Mastering Agentforce Debugging: Traces, Analytics, and Optimizer

  1. Q: Your agent is behaving oddly. How do you find out why?
  2. Follow-up Q: What is the difference between Agentforce Analytics and Agentforce Optimizer?
  3. Follow-up Q: How does Salesforce handle the cost of these Agentforce executions?

Agent Testing Strategy: How to Measure and Eliminate the AI Noise Floor

  1. Q: Your test suite drops from 91% to 89%. Is this a regression or just noise?
  2. Q: Does increasing the number of test cases reduce the noise floor?

The Ultimate Guide to Agentforce Testing Center & Custom Evaluations

  1. Q: What does Testing Center give you beyond manually typing messages into the Agent Builder preview?
  2. Q: How exactly does an "LLM Judge" evaluation work?
  3. Q: Can I integrate Testing Center into my DevOps process?

Agent Testing Strategy: Statistical Evaluation vs Traditional Testing

  1. Q: How do you reliably test an AI system that is not deterministic?

Mastering Salesforce Agentforce Escalation to Human Agents

  1. Q: What makes a high-quality human escalation in Agentforce?

Salesforce Agentforce Voice: Same Script, Different Physics

  1. Q: How does a voice agent fundamentally differ from a chat agent in Agentforce?

Salesforce Agentforce Channels: Where Do Agents Actually Live?

  1. Q: Which channels can an Agentforce agent run on?

Agentforce: How the Prompt Gets Built

  1. Q: How do you stop an agent from inventing a warranty or eligibility answer?
  2. Q1: The same question gives slightly different wording each time. Is that a bug?
  3. Q2: Where would you put a check that must run no matter what the model says?

Agentforce: The Reasoning Engine

  1. Q: Walk me through what happens when a customer sends a message to an agent.

Salesforce Einstein Trust Layer Explained: What It Actually Protects

  1. Q: What does the Einstein Trust Layer actually protect?
  2. Q: Why can't I find the Einstein Trust Layer settings in my Dev Org?

Agentforce Security Layers Explained: How Salesforce Protects AI Data

  1. Q: How is data protected in Agentforce?
  2. Q: What happens if an Apex action used by Agentforce is written using `without sharing`?
  3. Q: Does the Einstein Trust Layer mask data internally within the org?

Mastering Salesforce Agentforce Guardrails: Instructions vs. Filters

  1. Q: What is the exact difference between an instruction and a guardrail in Salesforce AI?

Mastering AI Agent Determinism: 4 Mechanisms for Bulletproof Guardrails

  1. Q: How do you stop an AI agent from doing something it shouldn't?
  2. Q1: Where does a guardrail fail to run, and what do you do about it?

Salesforce Agentforce: Why Retrieval Quality is the AI Ceiling

  1. Q: Your grounded Agentforce answers are wrong. Where is the very first place you look?
  2. Q: If my retrieval score is poor, what specific things should I change?
  3. Q: Self-Check: What are the four core steps of RAG architecture?
  4. Q: Self-Check: What does creating a Salesforce Data Library automatically build for you?

Agentforce Data Library: Search Index & Retrievers Explained

  1. Q: How would you set up document grounding for an Agentforce Agent?
  2. Q: Why are Metadata Filters on the Retriever so important?
  3. Q: Where do you turn on AI Citations so users can see where the Agent got its answer?

How Agentforce RAG Works: Chunking, Indexing & Retrievers in Salesforce

  1. Q: Explain exactly how an Agentforce agent answers a user question based on a PDF document.
  2. Q: Because vector search matches on meaning, are there any downsides?
  3. Q: What is a Data Library in Salesforce Data Cloud?
  4. Q: What is the difference between Vector Search and Hybrid Search in Salesforce?

🚨 Salesforce Prompt Builder: Avoiding the "Silent Blank" FLS Trap

  1. Q: What exactly happens if the running user cannot see a grounded field in a Prompt Template?
  2. Q: What are the primary grounding sources available in Prompt Builder?
  3. Q: How do you properly insert a merge field into a prompt template?
  4. Q: Why won't Debug Logs capture this specific missing field issue?
  5. Q: How can I audit my org to ensure FLS isn't breaking my AI templates?

🏗️ Agentforce Prompt Templates: Types, Writing & Best Practices

  1. Q: When should I use a Prompt Template instead of just putting the instructions directly inside my AI Agent?

🔍 Agentforce Grounding: The 5 Ways to Inject Real Salesforce Facts

  1. Q: Your Agentforce agent gives a highly confident but completely wrong answer to a customer. What is your exact first step?
  2. Q: What happens if I ground a prompt with a field, but the user running the agent doesn't have Field-Level Security (FLS) access to it?

🤖 Agentforce Routing: Transition vs. Delegation Explained

  1. Q: What is the exact difference between a transition and a delegation?
  2. Q1. If left unset in the configuration, what does agent_type default to?
  3. Q2. Can the after_reasoning block contain prompt text?
  4. Q3. Which orchestration method returns control back to the caller?

🚀 Mastering Agentforce Subagents: before_reasoning & after_reasoning Guardrails

  1. Q: Where would you put a business rule that must absolutely never be broken?
  2. Q: How do you decide between using the 'before' block versus the 'after' block?
  3. Q: Can I use AI prompt piping inside these logic blocks?

Agentforce Instructions: Understanding Logic Lines vs. Prompt Lines

  1. Q: What are the two kinds of reasoning instructions in Agentforce?
  2. Q: What happens if one specific subagent needs a different tone or rule than the rest of the system?
  3. Q: Does adding more detailed prompt lines make the AI perform better?

Agent Script: Understanding the 8 Essential Blocks in Salesforce Agentforce

  1. Q: What exactly is Agent Script, and how does it relate to the Agent Builder UI?

Agent Script: The Critical Difference Between Actions vs. Reasoning Actions in Agentforce

  1. Q: What is the primary difference between actions and reasoning.actions in Agentforce?
  2. Q: Can I use the same Apex action in both lists simultaneously?
  3. Q: What happens if I put too many tools in reasoning.actions?

Salesforce Agentforce Best Practices: Return Decisions, Not Raw Data to LLMs

  1. Q1: Why does your Apex action specifically return a "reason" string?
  2. Q2: Why is USER_MODE critical for AI-facing Apex?
  3. Q3: How do you handle bulkification when an AI agent calls the action?
  4. Q4: How can you test if your AI action is drifting?

Salesforce Agentforce: Writing Action Descriptions That Actually Work

  1. Q: How exactly does the Agentforce agent know which action to call?

Salesforce Agentforce Action Types: Flow, Apex, Prompt Templates, External Services, & MCP

  1. Q: When would you write an Agentforce action in Apex rather than Flow?

Salesforce Agentforce Architecture & Limits: The 15-Subagent Ceiling Explained

  1. Q: What are the primary limits in Agentforce, and which one dictates your technical architecture?

Mastering Agentforce Architecture: Hard Limits, Subagent Ceilings, and Scalable Design

  1. Q: What are the main Agentforce platform limits, and which one has the biggest impact on architecture?

Mastering Subagent Architecture in Salesforce Agentforce: Avoiding Description Overlaps and Ceiling Limits

  1. Q: Why design five narrow subagents instead of one comprehensive subagent?
  2. Q1: Your Agentforce deployment starts giving erratic or unexpected answers. How do you troubleshoot?
  3. Q2: Business stakeholders request a 16th subagent on a single agent. What is your architectural plan?

Salesforce Agentforce Routing & Subagent Architecture: Best Practices & Limits

  1. Q: How does Salesforce Agentforce determine which subagent should process a user message?

Why Shorter Instructions Win: Optimizing AI Agent Prompts for High Accuracy

  1. Q: Your agent routes correctly but gives inconsistent answers. What do you change first?

Agentforce: Agent Access vs Data Access

  1. Q: You built and activated an agent, but nobody can see it. What is wrong?

Agentforce: Which User the Agent Runs As

  1. Q: How do you prove an agent respects your security model?
  2. Q1: The agent returns an answer that is missing one field, with no error. What happened?
  3. Q2: Someone suggests giving the Service Agent user Modify All Data to save time. Your response?

Agentforce: Agent Types & Execution Identity

  1. Q: What is the difference between an Employee Agent and a Service Agent?
  2. Q1: Your Employee Agent works for you and fails for a support rep. Where do you look?
  3. Q2: How would you design permissions for a public-facing Service Agent handling outages?

Agentforce: Key Terms in Plain Words

  1. Q: What is the exact difference between an action and a tool?
  2. Q: How does Grounding actually prevent hallucinations?
  3. Q: What should I do if my Agent chooses the wrong Subagent?

Agentforce: Meridian PowerPack— Our Running Example for agentforce

  1. Q: Give me an example of a good Agentforce use case versus a bad one.
  2. Q: The business wants one single agent to handle outages, billing, tariffs, meter readings, complaints, and new property connections. How do you respond?
  3. Q: How does the agent interact with systems outside of Salesforce, like the legacy billing server?
  4. Q: What happens if a standard chatbot rule conflicts with Agentforce logic?

Agentforce: One Agent, Six Layers

  1. Q: Explain Agentforce to someone who knows Salesforce but has never built an agent.
  2. Q1: Where does Data 360 sit in that picture, and is it required?
  3. Q2: A stakeholder asks why you would not just build this in Flow. What do you say?

Claudeforce

Multi-Tool Workflows in Claude, MCP, and Salesforce Agentforce: Complete Orchestration Guide

  1. Q1: What is the single most important rule in a multi-tool workflow?
  2. Q2: The AI needs a customer's profile before making a recommendation. Which MCP primitive is relevant?
  3. Q3: When should Tools run in parallel?
  4. Q4: When should Tools run sequentially?
  5. Q5: Can Claude request several Tools in a single API response?
  6. Q6: What happens if one parallel Tool call fails while the others succeed?
  7. Q7: Is a multi-tool workflow automatically considered an AI Agent?
  8. Q8: Should we expose every available enterprise Tool to Claude at once?
  9. Q9: How do Salesforce governor limits and record locking affect parallel multi-tool calls?
  10. Q10: How can developers force Claude to call tools one at a time if parallel execution is unsafe for a specific workflow?

Master MCP Tool Calling: The Complete Lifecycle for Claude & Salesforce

  1. Q1: How does Claude know the exact field names and data types a tool expects?
  2. Q2: What happens if a Salesforce API or Flow fails halfway through a tool call?
  3. Q3: The AI needs to read a customer's static contract document before making a recommendation. Should we use a Tool, a Prompt, or a Resource?
  4. Q4: Leadership wants every executive Account Review generated by the AI to follow the exact same corporate structure. Which MCP primitive solves this?
  5. Q5: Can a single enterprise workflow combine Prompts, Resources, and Tools in the same turn?
  6. Q6: In Salesforce Hosted MCP Servers, when should I expose an Agentforce Agent vs. a Prompt Builder Template vs. an Apex/Flow Action as an MCP tool?
  7. Q7: Can an AI agent use an MCP Tool to view or edit Salesforce records that the logged-in user doesn't have access to?
  8. Q8: What is "Context Bloat" in MCP tool calling, and how does Salesforce prevent it?
  9. Q9: How do we stop an AI agent from accidentally calling a destructive tool—like deleting records or issuing a $50,000 refund? (Connecting the Dots)

Understanding MCP Primitives: Tools, Resources, and Prompts Explained

  1. Q1: What is the fastest way to remember the difference between Tools, Resources, and Prompts in an interview?
  2. Q2: Can a single enterprise AI workflow use all three MCP primitives at the exact same time?
  3. Q3: What is a Resource Template in MCP, and how is it different from a static Resource?
  4. Q4: Who actually executes the code when Claude calls an MCP Tool?
  5. Scenario 1: A sales rep types, "Create a follow-up task for ABC Healthcare due tomorrow." Which MCP primitive handles this?
  6. Scenario 2: Before drafting a cold email, the AI needs to read your company's official "Compliance & Tone Guidelines" PDF. Which primitive fits best?
  7. Scenario 3: Your VP of Sales demands that every AI-generated deal summary follows the exact same 4-bullet MEDDIC format. Which primitive solves this?
  8. Scenario 4: How does MCP connect with Salesforce Agentforce in a real enterprise architecture?
  9. Scenario 5: From a cybersecurity perspective, why is separating Tools from Resources such a big deal?
  10. Scenario 6: When both a Resource and a Tool can read data from Salesforce, how do I choose which one to build?

Tokens, Context Windows & Prompts: How AI Understands Text

  1. Q1: Why does an AI chatbot suddenly "forget" details I told it 20 minutes ago in the same chat?
  2. Q2: If Claude already understands human language and business concepts so well, why does a "Claudeforce" setup need Salesforce data?
  3. Scenario 1: Since modern Claude models support 200K to 1M tokens, shouldn't I just dump all 500 related Case records into my prompt instead of filtering them?
  4. Scenario 2: How can I quickly estimate how many tokens my Salesforce Prompt Builder template will consume?
  5. Scenario 3: How does Salesforce Agentforce dynamically inject the right CRM data into the prompt at runtime?
  6. Scenario 4: What is Prompt Caching, and how does it cut enterprise AI costs by up to 90%?
  7. Scenario 5: What happens if a user's prompt plus the grounded CRM records exceed the model's maximum context window?
  8. Scenario 6: Why do AI providers set a separate max_tokens output limit much smaller than the total context window?

How Claude, MCP, and Salesforce Work Together: A Complete Guide

  1. Q1: Where exactly does Salesforce fit into the Claude + MCP architecture?
  2. Q2: Does MCP replace Salesforce REST APIs, GraphQL, or SOQL?
  3. Q3: Does Claude automatically get access to all of our company's Salesforce data?
  4. Q4: What happens when a user asks Claude to update or modify an Opportunity?
  5. Q5: How does this differ from Salesforce's native Agentforce platform?
  6. Q6: Does connecting via MCP consume Salesforce API Governor Limits?
  7. Q7: How should we set up authentication between an MCP Server and Salesforce in 2026?

Master MCP Security & Governance for Enterprise AI

  1. Q1: What is the exact difference between authentication and authorization when connecting an AI agent?
  2. Q2: Why does Salesforce require an External Client App (ECA) instead of a classic Connected App for Hosted MCP Servers?
  3. Q3: Should an internal AI agent ever be given System Administrator access to save setup time?
  4. Q4: Why are audit logs so critical when AI agents use MCP tools?
  5. Q5: How do you guarantee Claude never leaks sensitive executive Salesforce data to the wrong employee?
  6. Q6: What is "token passthrough," and why does the MCP specification ban it?
  7. Q7: What if a user has edit rights in Salesforce, but I only want their AI agent to have read-only access?
  8. Q8: What are "prompt injection" and "tool poisoning," and how do we defend against them in MCP?
  9. Q9: What happens if an employee leaves the company while their AI client still has an active refresh token?
  10. Q10: Connecting the Dots — How does mastering MCP security prepare you for full Claudeforce and Agentforce architectures?

🚀 MCP Architecture Explained: Connecting Claude AI to Salesforce

  1. Q: What is the exact difference between an MCP Client and an MCP Server?
  2. Q: Can one AI Host connect to multiple MCP Servers at the same time?
  3. Q: Does the MCP Server decide who gets access to the data?
  4. Q: Follow-up: How do you build an MCP Server for Salesforce?

The Ultimate Guide to Model Context Protocol (MCP): The USB-C for AI

  1. Q: Is MCP the same thing as a REST API?
  2. Q: Does MCP replace existing APIs like Salesforce's?
  3. Q: Why does MCP matter so much for AI agents?
  4. Q: Who created MCP, and is it proprietary?
  5. Q: Does Salesforce offer its own MCP servers?
  6. Q: What's the difference between a local and a remote MCP server?
  7. Q: How is MCP different from regular tool use?
  8. Q: Is it safe to connect any MCP server I find online?

AI Agents vs. AI Assistants: The Claudeforce Evolution

  1. Q: Is every chatbot essentially an AI Agent?
  2. Q: Why are "Tools" so critical for AI Agents to function in Salesforce?
  3. Q: How does MCP (Model Context Protocol) fit into the Claudeforce architecture?

The Complete Guide to Claude Tool Use: Connecting AI to Salesforce

  1. Q: Can Claude directly update Salesforce records?
  2. Q: How does Claude know which tool to use?
  3. Q: What happens if a tool fails or returns an error?
  4. Q: Can Claude use more than one tool in a single request?
  5. Q: How is tool use different from MCP?
  6. Q: How do you keep "write" tools safe?
  7. Q: What makes a good tool description?

Claude Reasoning: How Claude Understands, Analyzes, and Solves Problems

  1. Q: Is Claude actually "thinking" like a human?
  2. Q: Why does Claude need Salesforce data to reason well?
  3. Q: What is Claudeforce?
  4. Q: Does more thinking always give a better answer?
  5. Q: How is reasoning different from tool use?
  6. Q: How do you check whether Claude's reasoning is right?

Mastering Claude Model Families: Opus vs. Sonnet vs. Haiku

  1. Q: Should every Salesforce user default to the most powerful Claude model (Opus)?
  2. Q: Is Opus always smarter than Sonnet?
  3. Q: Why not just use the fastest model (Haiku) for everything to save money?

What is Claude AI? Understanding the Intelligence Behind Claudeforce

  1. Q1: A manager asks Claude, "Show me my top five opportunities." Why can't Claude answer this from its base knowledge?
  2. Q2: If Claude is so smart and can reason, why do we still need standard Salesforce business rules and workflows?
  3. Q3: Should our development team always default to using the most powerful Claude model (like Opus)?
  4. Q4: How do we keep our Salesforce data secure when sending it to Anthropic's Claude?

Deterministic vs Probabilistic AI: Balancing Rules and Reasoning in Salesforce

  1. Q: Why not just replace all our old business rules and Flows with AI Agents?
  2. Q: Should a Large Language Model directly update enterprise database records using SQL or SOQL?
  3. Q: How do you perform QA (Quality Assurance) on a probabilistic system if the answers keep changing?
  4. Q: How does Salesforce Agentforce specifically handle the handoff between probabilistic and deterministic layers?
  5. Follow-up Question: How do we prevent the probabilistic model (the LLM) from accidentally leaking sensitive customer data that was fetched by a deterministic tool?

🤖 AI Hallucinations Explained: How Salesforce Enterprise AI Fixes Bad Answers

  1. Q: Why exactly can AI give wrong answers if it is so smart?
  2. Q: Why use "grounding" instead of just retraining Claude or ChatGPT with our company data?
  3. Q: What is RAG, and how does it fit into this?

AI Basics: Training vs. Inference Explained (With Salesforce Examples)

  1. Q: Does an LLM like Claude get permanently smarter after every question I ask?
  2. Q: If models aren't retrained all the time, how does Salesforce Agentforce know about my company's live data?
  3. Q: When would you fine-tune a model instead of using RAG?
  4. Q: What is a "knowledge cutoff," and how does it relate to training?
  5. Q: Why is AI usage usually priced by tokens?
  6. Q: If the weights are frozen, why can the same prompt give different answers?

🧠 How LLMs Generate Answers: Inside Claude & Salesforce Agentforce

  1. Q: Does an LLM store a copy of every answer it gives?
  2. Q: Why do LLMs sometimes give incorrect answers (hallucinate)?

What Is an LLM (Large Language Model) in Salesforce? | Complete AI Guide

  1. Q: Is an LLM just a highly advanced database?
  2. Q: Why does Claude need Salesforce data if the LLM is already so smart?
  3. Q: What is "Claudeforce" and how does Salesforce keep my data secure when using it?

What is Generative AI? The Ultimate Guide for Salesforce & Claude Users

  1. Q: Is Generative AI just a fancier Google Search?
  2. Q: Why can't a public AI model (like standard ChatGPT or Claude) automatically answer questions about my company's specific pipeline?
  3. Q: What is an AI "Agent" compared to Generative AI?