All 61 questions
How would you build an AI agent, and where would frameworks like LangGraph or MCP fit?
Asked in 7 of 181 interviews · last asked Sep 10 · Design · Mid-level
Show answer
I would define the task, choose a model and tools, and build a loop that selects actions, checks results, and stops. I would track state, limit tool permissions, and test failure cases. LangGraph can manage state, branching, and pauses for human approval. MCP is a protocol, not an agent framework; it standardizes connections to tools and data sources.
- Start with a simple workflow; use agent decisions only where needed.
- Set time, cost, and step limits, with clear error handling.
- Require human approval for risky actions and test task success.
- LangGraph manages execution; MCP connects tools and data. Neither is required.
What the interviewer checks: The interviewer checks whether you can design a safe, testable agent and distinguish orchestration from tool integration.
Other ways it was asked
- “How would you build an AI agent?”
- “What components are needed to build an AI agent?”
- “How would you design the high-level architecture of an AI agent?”
When would you choose LangGraph over LangChain?
Asked in 5 of 181 interviews · last asked Sep 27 · Concept · Senior
Show answer
I would choose LangGraph when I need explicit control over a stateful workflow, including branches, loops, and human approval steps. I would use LangChain for a faster start with standard agents, model integrations, and tools. They work together: LangChain’s current agent API is built on LangGraph.
- LangGraph lets you define steps, transitions, and shared state explicitly.
- Checkpoints let workflows resume from saved state, including after human review.
- LangChain agents already use LangGraph; choose LangGraph directly when you need more control.
What the interviewer checks: The interviewer is checking whether you can match workflow complexity to the right level of control.
Other ways it was asked
- “What is the difference between LangChain and LangGraph?”
- “When would you use LangChain, and when would you use LangGraph?”
- “When would you choose LangGraph over LangChain for an LLM application?”
How does the ReAct agent pattern work, and how does it differ from plan-and-execute?
Asked in 2 of 181 interviews · last asked Jul 22 · Concept · Mid-level
Show answer
ReAct repeats a loop of reasoning, taking an action, and using the result to choose the next step. Plan-and-execute creates a plan first, then carries out its steps, with replanning when needed. ReAct adapts step by step, while plan-and-execute gives longer tasks more structure.
- ReAct uses tool results as feedback before choosing its next action.
- Plan-and-execute separates planning from execution; the plan does not have to stay fixed.
- ReAct can lose direction over many steps; an upfront plan can become outdated.
What the interviewer checks: The interviewer is checking whether you understand both control flows and when each is useful.
Other ways it was asked
- “What can you tell me about the ReAct pattern in the design of agentic systems?”
- “How do ReAct and plan-and-execute agent patterns differ?”
When would you use a fixed workflow, a single agent, or a multi-agent system?
Asked in 2 of 181 interviews · last asked Aug 13 · Concept · Senior
Show answer
I use a fixed workflow when the steps are known and predictable. I use a single agent when the task needs flexible decisions about steps or tools. I use multiple agents when separate roles or parallel work provide clear value. I start with the simplest option because more agents add cost and coordination risks.
- Fixed workflows are easier to test and control; they can still include AI calls.
- A single agent can choose tools and adapt its next step.
- Multiple agents suit separable tasks, but add delays, cost, and coordination failures.
- Use evaluations to check whether added complexity improves results.
What the interviewer checks: The interviewer is checking whether you match the design to the task rather than assuming more agents are better.
Other ways it was asked
- “Why might you build a fixed workflow when agent technology is available?”
How would you design a production multi-agent system end to end?
Asked in 2 of 181 interviews · last asked Oct 7 · Design · Senior
Show answer
I would start with the task, success metrics, and whether multiple agents are needed. I would give each agent a clear role and use an orchestrator to route work through explicit steps. I would define shared state, message formats, and tool permissions. Before launch, I would test failures, add monitoring and human approval, and roll out gradually.
- Use sequential steps for dependencies and parallel agents for independent work; define how conflicting results are resolved.
- Persist workflow state; use timeouts, bounded retries, and duplicate protection for actions with side effects.
- Limit tool access, validate outputs, and require human approval for high-risk actions.
- Evaluate end-to-end quality, trace agent decisions, and enforce cost, latency, and step limits.
What the interviewer checks: The interviewer is checking whether you can turn agent roles into a reliable, secure, and measurable production workflow.
Other ways it was asked
- “How would you orchestrate a multi-agent system?”
What information would you store in a LangGraph state object?
Asked once · last asked Jul 20 · Design · Mid-level
Other ways it was asked
- “How does LangGraph pass information between nodes, and what does the state look like?”
If you want to set up a new multi-agent assistant system for question answering and task execution, what tools would you use across the toolchain?
Asked once · last asked Oct 2 · Design · Senior
Other ways it was asked
- “What tools would you use across the toolchain to build a multi-agent assistant system?”
If you were designing an enterprise-grade agentic system from scratch today, what major architectural decisions would you need to make? Talk me through how you would approach the problem and the aspects you would cover.
Asked once · last asked Apr 17 · Design · Senior
How do Google ADK and the OpenAI Agent SDK compare for building agentic systems?
Asked once · last asked Apr 17 · Concept · Senior
How would you design an agentic AI workflow with LangChain, CrewAI, or a similar framework, covering its use case, orchestration, tool usage, memory, and reliability?
Asked once · last asked May 27 · Design · Senior
Other ways it was asked
- “Describe an agentic AI workflow you built with LangChain, CrewAI, or a similar framework, including its use case, orchestration, tool usage, memory, and reliability.”
What is the Agent node in n8n?
Asked once · last asked Jun 17 · Tooling · Mid-level
What are several design patterns for building multi-step agentic systems?
Asked once · last asked Jun 23 · Design · Senior
How do you choose between a low-code option and custom code for a node in an AI agent?
Asked once · last asked Jul 16 · Scenario · Senior
If you had to build an agentic AI system, what would be your plan? How would you design this system?
Asked once · last asked Jul 16 · Design · Senior
What is reflection in agentic AI?
Asked once · last asked Jul 16 · Concept · Mid-level
What are self-correction and refinement in agentic AI?
Asked once · last asked Jul 16 · Concept · Mid-level
What might a LangGraph flow look like, and why choose LangGraph over plain tool calling?
Asked once · last asked Jul 20 · Design · Mid-level
What nodes and edges would you define in a LangGraph workflow?
Asked once · last asked Jul 20 · Design · Mid-level
How would you design a multi-agent system with one agent retrieving data, one storing and managing data, and one refining the final response?
Asked once · last asked Jul 25 · Design · Senior
Other ways it was asked
- “If you need to design a multi-agent system with one agent retrieving data, one storing/managing data, and one fine-tuning the final response, what would your design approach be?”
How would you build a tool-using chat agent like Claude, and which libraries could speed up development?
Asked once · last asked Jul 28 · Design · Senior
How would you build or optimize an LLM-powered or agentic application, including prompt design, tool use, and orchestration?
Asked once · last asked Aug 2 · Design · Senior
Other ways it was asked
- “How have you approached building or optimizing LLM-powered or agentic applications? Walk me through one example, including how you handled prompt design, tool use, or orchestration.”
How would you architect a complex agentic or LLM system end to end, and address its hardest engineering challenges?
Asked once · last asked Aug 2 · Design · Senior
Other ways it was asked
- “What is the most complex agentic or LLM system you have built end to end? Walk me through its architecture and the hardest engineering problem you had to solve.”
How would you define state in a LangGraph workflow and handle steps that fail or produce invalid output?
Asked once · last asked Aug 2 · Scenario · Senior
In CrewAI, how do agents and tasks differ, and where should you assign a research topic?
Asked once · last asked Aug 7 · Concept · Junior
Other ways it was asked
- “In CrewAI, what is the difference between an agent and a task?”
- “In CrewAI, should a research topic be assigned to a task or to an agent?”
Would you use LangGraph's default ReAct agent or build a custom agent?
Asked once · last asked Aug 7 · Design · Senior
If the access-granting system is a black box, would the ticket-handling agent still need its own database?
Asked once · last asked Aug 7 · Design · Senior
How do you coordinate cross-review among AI agents, and what tools support that process?
Asked once · last asked Aug 10 · Scenario · Senior
How would you validate tool outputs and enforce stop conditions in a LangGraph workflow, including the checks or guards you would implement?
Asked once · last asked Aug 16 · Tooling · Mid-level
Other ways it was asked
- “How exactly did you validate tool outputs and enforce stop conditions in your LangGraph flow? What checks or guards were in place?”
Why would you choose a multi-agent architecture instead of a single LLM call?
Asked once · last asked Aug 18 · Scenario · Mid-level
Other ways it was asked
- “Why did you use the multi-agent architecture instead of just using a single LLM call?”
- “When might a multi-agent architecture add unnecessary complexity compared with a single LLM call?”
How would you design a seven-agent architecture, and what purpose would each node serve?
Asked once · last asked Aug 18 · Design · Mid-level
What failure modes can occur in parallel agent orchestration?
Asked once · last asked Aug 18 · Scenario · Mid-level
How would you separate an agent/tool layer, a storage layer, and a serving or dashboard layer in a system architecture?
Asked once · last asked Aug 19 · Design · Senior
How do you define the boundary between agents when there are 150 applications? Does each agent touch all the data, or do you segment them by system or function?
Asked once · last asked Aug 19 · Design · Senior
In an agent framework such as Google ADK, how would you implement a function that represents a state or step in the workflow?
Asked once · last asked Aug 20 · Tooling · Mid-level
In Claude Cowork, how are state management and function calling handled?
Asked once · last asked Aug 20 · Tooling · Mid-level
Beyond state management, what high-level architectural components would you define for an agent orchestration harness?
Asked once · last asked Aug 20 · Design · Senior
How would you choose between one shared internal AI assistant and multiple team-specific assistants?
Asked once · last asked Aug 22 · Design · Senior
Other ways it was asked
- “What are the trade-offs between one shared internal AI assistant and multiple assistants tailored to different teams?”
- “What factors influenced the choice between one shared assistant and multiple team-specific assistants?”
Could you describe the system interface or harness that wraps around the model?
Asked once · last asked Aug 22 · Design · Senior
Walk through decomposing a complex workflow into sequential agent subtasks, executing them reliably, and combining the results.
Asked once · last asked Aug 26 · Design · Senior
Other ways it was asked
- “How would you decompose an ambiguous, complex workflow into sequential deterministic subtasks for autonomous agents, execute them reliably, and integrate the results?”
- “Can you describe a specific example of decomposing a complex problem into sequential subtasks for AI agents?”
When would you use a sequence of LLM calls versus one agent with multiple roles that iterates over user input?
Asked once · last asked Sep 2 · Design · Senior
Can you explain the complete agent loop and how the model decides when to invoke a tool like Internet search?
Asked once · last asked Sep 22 · Concept · Senior
How would you implement an intent classifier for routing requests in an agentic LLM system: as a separate machine learning model or as an LLM prompt?
Asked once · last asked Sep 22 · Scenario · Senior
Other ways it was asked
- “What intent categories would you define for a classifier that routes requests in an agentic LLM system?”
- “What intent categories would you define for an agentic LLM routing classifier, and how would you measure its classification accuracy?”
How would you compose an agentic AI orchestrator where a Python service sends a prompt to an LLM, interprets response codes, and conditionally calls two external tools, while handling tool failures and maintaining transactional consistency as requirements evolve?
Asked once · last asked Sep 24 · Design · Senior
How can you set up multiple coding agents in Claude?
Asked once · last asked Sep 24 · Tooling · Mid-level
What types of agents can you build with LangGraph?
Asked once · last asked Sep 25 · Concept · Mid-level
How do you implement a retry and a human approval node in LangGraph?
Asked once · last asked Sep 27 · Tooling · Senior
How would you implement a multi-agent assistant system, including using AI coding agents or harnesses?
Asked once · last asked Oct 2 · Design · Senior
What technologies would you use to build RAG systems and agentic workflows?
Asked once · last asked Oct 5 · Concept · Senior
What would you use to enable agent-to-agent interactions?
Asked once · last asked Oct 8 · Design · Senior
How would you design a multi-agent application, define each agent's responsibilities, and coordinate their interactions?
Asked once · last asked Oct 9 · Design · Senior
What is scaffolding in LLM application development, and why is it used?
Asked once · last asked Feb 5 · Concept · Mid-level
Why would you use LangChain to build an AI application?
Asked once · last asked Apr 29 · Tooling · Mid-level
What agentic AI platforms are available in the market?
Asked once · last asked Jul 17 · Concept · Mid-level
What are the different types of tool-using agents?
Asked once · last asked Jul 25 · Concept · Junior
What inputs should you pass to a planner agent?
Asked once · last asked Jul 31 · Concept · Mid-level
Can you explain LangGraph?
Asked once · last asked Aug 2 · Concept · Mid-level
In simple terms, in 20 to 50 words, what is CrewAI?
Asked once · last asked Aug 7 · Concept · Junior
What is an agent harness?
Asked once · last asked Aug 20 · Concept · Mid-level
What is LangChain?
Asked once · last asked Sep 21 · Concept · Junior
What do you mean by a multi-agent AI platform?
Asked once · last asked Oct 9 · Concept · Mid-level
How would you justify your choice of approach when designing a multi-agent AI platform?
Asked once · last asked Oct 9 · Design · Senior
More Agentic AI interview questions
- All Agentic AI interview questions
- Agentic AI Basics and Use Case Interview Questions
- AI Agent Tool Calling and MCP Interview Questions
- AI Agent Memory, Context Management, and RAG Interview Questions
- AI Agent System Design Interview Questions
- AI Agent Production Deployment and Reliability Interview Questions
- AI Agent Security, Guardrails, and Governance Interview Questions
How this list is made
Every question here was asked by a real interviewer in a real job interview that a candidate ran MeetAssist in. After the call, the interviewer's questions are pulled from the transcript. Practice sessions, tests, exams and anything that was not a job interview are dropped.
Questions are rewritten so they make sense on their own: names of people, employers and clients are removed, and speech-to-text mistakes are fixed. Small talk, logistics and questions about one candidate's own résumé are left out. Rephrasings of the same question are merged, and the count is the number of different interviews it came up in.
The interviews come from candidates who use MeetAssist, so they lean towards software, data and AI roles: the shares describe these interviews, not every Agentic AI interview. Answers to the most asked questions are written and checked by an editor; a "common mistake" is shown only when reviews of real answers in these interviews show it, with the count.
181 interviews, 116 candidates, Nov 2025 to Oct 2026. Updated every month with new interviews.
Or add the Chrome extension