AI Agent Interview Prep

AI Agents Interview Topics

• Agent vs LLM call -- LLM: one shot Q&A; agent: iterative loop with tool use and memory
• ReAct pattern -- Reason (thought) + Act (tool call) + Observe (result) loop until done
• Tool definition -- name, description, and JSON schema; description quality determines selection accuracy
• Agent memory types -- short-term (conversation), long-term (vector store), entity, episodic
• Multi-agent benefits -- specialisation, parallelism, independent verification of outputs
• HITL -- pause agent before destructive actions; LangGraph checkpointing enables pause/resume
• Agent safety -- sanitise tool outputs (injection), cap iterations, audit log, least privilege tools
• Evaluation -- task success rate, avg turns, tool accuracy, hallucination rate, cost per task
• LangGraph vs LangChain -- LangGraph for stateful cyclic workflows; LangChain for linear chains
• Code execution safety -- always sandbox: subprocess timeout, Docker with no network, resource limits