Core concepts and architecture
- What is an AI agent, and how is it different from a chatbot?
- Explain the difference between an LLM and an agent framework
- What is the agent loop, and what are its core steps?
- What is function/tool calling and how does the model decide which tool to call?
- Explain the difference between JSON tool-calling and code-execution agents (e.g. smolagents' CodeAgent)
- What is a handoff, and when would you use one over a single agent with more tools?
- What is the Model Context Protocol (MCP), and what problem does it solve?
- How does the Agent2Agent (A2A) protocol differ from MCP?
- What is RunContext / dependency injection in an agent framework, and why does it matter?
- Explain structured output and why it matters for agent reliability
Memory, RAG & data
- What's the difference between short-term and long-term agent memory?
- Explain a basic RAG pipeline end to end
- What is chunking, and why does chunking strategy affect retrieval quality?
- What is hybrid search, and when would you combine it with vector similarity?
- When would you reach for pgvector instead of a dedicated vector database?
- Explain prompt caching and why it matters for cost at scale
Execution, safety & production
- Why does an agent that executes code need a sandbox?
- What's the difference between a sandbox with persistent storage and one without?
- What is operational misalignment, and how is it different from a normal bug?
- What's the minimum viable audit trail for a production agent?
- Explain the difference between agent-native and LLM-first observability tools
- What are evals, and how are they different from unit tests?
- How would you estimate and control the token cost of an agent running in production?
- What's the difference between a gateway/router and a direct model provider integration?
- Name three ways to reduce an agent's dependency on calling an LLM for every step
Why it matters
Agent roles mix backend, ML ops and security, so interviews test breadth. A clear story about one design choice beats a list of framework names.
Where people get stuck
- Saying "we just used LangChain" with no reasoning. Interviewers will ask why.
- Describing tool access without mentioning safety. That's a red flag for enterprise teams.
- No numbers. Production stories need latency, cost and success rate.
Try this
Pick five questions from this page, answer each in writing with a diagram and practise saying it in under two minutes.
Interview question
Walk me through a production agent architecture.
User channel, runner service, model gateway, tool layer (MCP or APIs), memory and RAG stores, a sandbox for code, observability and evals, then deployment and cost controls. Go deep on the piece closest to the role.
Resources
YouTube learning
Core concepts from video walkthroughs:


