Q1.What does "tool calling" (function calling) allow an LLM to do?
Q2.In the ReAct pattern, what do the "Reason" and "Act" steps refer to?
Q3.Who actually executes a tool call requested by an LLM?
Q4.What is typically included in a tool's definition given to an LLM?
Q5.Why might an agent loop include a maximum step limit?
Q6.What is a common risk when an agent's tool includes untrusted external content (e.g. a webpage) in its output?
Q7.What is "planning" in the context of LLM agents?
Q8.Why are agentic systems generally more expensive and slower than single-turn LLM calls?
Q9.What is a common technique for giving an agent "memory" beyond a single context window?
Q10.What does "human-in-the-loop" mean for an agent system?
Q11.What distinguishes a "chatbot" from an "agent" in this context?
Q12.What is a typical structured format for tool call outputs from modern LLM APIs?
Q13.Why should tool outputs be validated before being fed back to the model?
Q14.What is a common failure mode of multi-step agent loops?
Q15.What is meant by "orchestration" in multi-agent systems?
Q16.Why might an agent system use a "router" step before invoking a specific agent or tool?
Q17.What is a key difference between giving an LLM a single powerful tool versus many narrow tools?
Q18.What does "self-correction" or "reflection" refer to in agent design?
Q19.Why is timeout/retry handling important when an agent calls external tools/APIs?
Q20.What is a common way to evaluate whether an agent successfully completed a task?
Q21.What risk does giving an agent a tool with write/destructive access (e.g. delete a file, send an email) introduce?
Q22.What is the purpose of few-shot examples of tool use in an agent's system prompt?
Q23.In an agent loop, what typically happens immediately after a tool call is executed?
Q24.Why might an agent system impose a strict allowlist of permitted tools/actions rather than giving the model unrestricted capability?
Q25.What's a key difference between a single-agent system with tools and a multi-agent system?
