Home▸Free interview questions▸Agentic AI interview questions
free · no sign-up · no loginAgentic AI interview questions, answered
20 questions with short answers, picked from the Q&A in the AI Agent Developer course. Tap a question to see the answer.
What is agentic AI, and how is it different from generative AI?
Agentic AI is an AI system that works toward a goal by itself. It plans the steps, calls tools (search, APIs, code, databases), looks at the results and keeps going until the task is done, inside limits you set. Generative AI answers one prompt with one output. Agentic AI runs a loop: plan, act, check, repeat.
- Plan: break the goal into steps
- Act: call a tool or another agent
- Check: read the result, fix mistakes
- Stop: finish, or hand over to a person
No question matches that search. Try a shorter word.
Agentic AI interview questions
The full set, with labs and mock interviews, is in the AI Agent Developer course.
Agents & Multi-Agent
1What is an AI agent?
An AI agent is a model with tools, working in a loop toward a goal. It decides a step, does it, looks at the result, and repeats until done or it needs a human.
2What is ReAct?
ReAct means Reason, Act, Observe, repeat. The model thinks, uses a tool, reads the result, and thinks again.
3Workflow or agent?
In a workflow, you fix the steps in code. In an agent, the model chooses the steps. Start with a workflow, and use an agent only when the steps can't be planned.
4How do agents keep state between steps?
Mainly through the message history: every tool call and result is added to it. For longer work they also save state outside the model, in files, a database or checkpoints, so they can continue later or after a crash.
5How should an agent handle an unclear request?
If a wrong guess would be costly, ask one short clarifying question. If the risk is low, make a sensible assumption, say what it assumed, and carry on.
6What is an 'agentic workflow'?
A workflow where the overall path is set in code, but some steps use an LLM, often with tools. It sits between a simple prompt and a fully autonomous agent: predictable, but still flexible where needed.
7What is routing in agent design?
A first step reads the request and sends it to the right handler, like refunds, tech support or sales, each with its own prompt and tools. Every handler stays small and focused, so it's easier to build and test.
8Explain parallelization: sectioning and voting.
Sectioning splits a task into independent parts that run at the same time, like checking safety while answering. Voting runs the same task several times and takes the majority answer for more confidence.
9Stateless vs stateful agents: trade-offs?
Stateless agents start fresh every time, so they are simple to scale. Stateful agents remember and save progress, which helps long tasks but needs storage.
10What are Agent Skills?
Packaged folders of instructions, scripts and files for one kind of job, like making slides or filling forms. The agent sees only short descriptions, and loads a skill's full contents only when the task needs it.
11Why does an agent need checkpointing?
Checkpointing saves progress after each step. If it crashes, it continues without repeating steps, like paying a refund twice. It also helps with debugging and with waiting for a human's approval.
12What limits should every agent have?
A maximum number of steps, a cost limit, a time limit, and a check for repeating the same call. When a limit is hit, stop and hand over with a summary.
13When do you use multiple agents?
Use more than one agent only when one agent can't handle the task, or parts can run at the same time. Otherwise one agent is cheaper and easier to fix.
14When should you NOT use an agent?
When the steps are known and fixed, when one model call is enough, when mistakes are costly and can't be checked, or when speed and cost matter most. A simple workflow or a single prompt is more reliable in those cases.
15How do you build an agent loop without a framework?
Keep a list of messages. Call the model with your tools. If it asks for a tool, run it, add the result to the messages, and call the model again. Stop when it gives a final answer or hits a step or cost limit. That's the whole loop.
16Name the main multi-agent patterns.
A manager agent with worker agents, a step-by-step chain, many agents in parallel, a reviewer that checks another agent, and handing a task from one agent to another.
17What are the downsides of multi-agent systems?
They use many more tokens, take longer, can confuse each other, and are more difficult to debug. Keep their messages structured and trace every step.
18How do you keep humans in control of risky actions?
Pause before anything that can't be undone. Show a human the evidence. Continue only when they approve.
19How big should each agent step be?
Big enough to make real progress, small enough to check. Steps that are too small waste calls and tokens; steps that are too big hide errors. A good step ends with something you can verify, like a passing test.
Agentic AI System Design: 20 Scenario Questions
20Design an agent that can recover from failures.
Save a checkpoint after every step, so a crash does not lose the work. Make each tool call safe to repeat. Retry a failed step a limited number of times with a wait in between. If it still fails, try another route or hand over to a human with a summary.
More free interview questions by topic
All 91 free questionsRAGMCPLLMAI system designLLM evaluation and guardrailsFDE
Now test yourself
Play the free quiz, or try a mock interview where you answer out loud and get a score.