AI Interview Questions

Introduction

Curated interview questions covering beginner → expert → scenario. Read aloud and answer in < 2 minutes each — that's the bar.

Understanding the topic

Core concepts:

• Beginner: What is Agentic AI? Tool calling? RAG?
• Intermediate: Compare ReAct vs Plan-Execute. Cost trade-offs?
• Advanced: Design a multi-agent code reviewer with safety.
• Scenario: Latency p95 exploded after a prompt change. Triage.
• Debugging: Agent hallucinates a tool name. Fix.
• Architecture: Choose a vector DB at 1B rows. Justify.
• Safety: Prompt injection arrives from a web tool result. Defend.
• Cost: Halve LLM bill without quality regression. How?
• Eval: Build a golden eval set for a customer-support agent.
• Org: How do you ship prompts safely across 50 engineers?

Syntax reference

Visual workflow / architecture:

┌──────────────┐
│   Exercise   │
└──────┬───────┘
       ▼
  Build · Test · Eval
       │
       ▼
  Compare with model answer

Real-world use

These are real questions reported from Anthropic, OpenAI, Cursor, Cognition, Glean, Notion AI interviews.

Best practices

• Practice OUT LOUD.
• Bring a one-page diagram to every architecture answer.

Common mistakes

• Memorising answers — interviewers probe; show first-principles.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Define Agentic AI in 30 seconds.
• Q2. Compare ReAct and Plan-Execute briefly.
• Q3. Scenario: cost doubled — first 3 questions you ask.