Graded Quiz: Build Self-Improving Agents with LangGraph :Agentic AI with LangChain and LangGraph (BM RAG and Agentic AI Professional Certificate) Answers 2025
Question 1
Why are utility-based agents more advanced?
❌ They select any action that completes a goal
❌ They update behavior via trial-and-error
❌ They avoid internal models
✅ They evaluate and rank outcomes using a utility function before selecting an action
Explanation:
Utility-based agents assess trade-offs and choose the best action, not just any goal-satisfying one.
Question 2
Advantage of a reflection loop:
✅ It allows the AI to self-critique and iteratively improve responses
❌ Removes human oversight
❌ Prevents incorrect facts completely
❌ Reduces API calls
Explanation:
Reflection improves quality via critique → revision cycles.
Question 3
Next step in Reflexion workflow (“foods high in iron”):
❌ Generate new random response
❌ Return initial answer
❌ Start a new agent
✅ Send response to a tool for citations, then revisor updates answer with source-backed information
Explanation:
Reflexion improves reliability by combining tools + revision steps.
Question 4
Improving health advice with citations:
❌ Manually add citations
✅ Pass both initial output + tool data to the revisor for a citation-backed revision
❌ Restart responder
❌ Remove constraints
Explanation:
Revisor integrates authoritative references and corrects unsafe advice.
Question 5
Purpose of tool_calls in AIMessage:
❌ Log queries
❌ Run revisor’s system message
❌ Define entry/exit nodes
✅ Store key-value outputs like answer, reflection, search queries
Explanation:tool_calls holds structured tool outputs needed for the Reflexion cycle.
Question 6
Correct first step in ReAct:
✅ Think: “I need to check Zurich’s weather”
❌ Recommend clothing immediately
❌ Ask for closet inventory
❌ Query clothing tool first
Explanation:
ReAct always begins with THINK → ACT → OBSERVE.
Question 7
Purpose of the “Observation” step:
❌ Log user history
✅ Show the tool’s returned result during graph execution
❌ Display final answer
❌ Evaluate reasoning
Explanation:
Observation = tool output, which the LLM uses for the next reasoning step.
🧾 Summary Table
| Q No. | Correct Answer | Key Concept |
|---|---|---|
| 1 | Utility evaluates ranked outcomes | Utility-based agents |
| 2 | Reflection improves responses | Reflection loop |
| 3 | Tool fetches citations + revise | Reflexion workflow |
| 4 | Pass outputs + tools → revisor | Citation-based revision |
| 5 | Stores tool outputs | AIMessage.tool_calls |
| 6 | Think first | ReAct pattern |
| 7 | Tool output shown for next reasoning | Observation step |