Test your understanding of AI agents, tools, threads, runs, multi-agent patterns, and safety.
Question 1: Your team needs a chatbot that answers customer questions about uploaded product manuals. Which agent tool is most appropriate?
A) Code Interpreter B) File Search C) Bing Search D) Function Calling
Answer: B) File Search – File Search provides vector-based retrieval over uploaded documents, making it ideal for Q&A over product manuals.
Question 2: You create a run and its status is requires_action. What does this mean?
A) The agent has finished processing B) The agent encountered an error and needs a restart C) The agent is waiting for you to execute a function call and return results D) The thread has expired and needs to be recreated
Answer: C) – requires_action means the agent invoked a function call tool and is waiting for your code to execute the function and submit the output via submit_tool_outputs_to_run.
Question 3: What is the correct order of the agent run lifecycle for a successful execution?
A) in_progress -> queued -> completed B) queued -> completed -> in_progress C) queued -> in_progress -> completed D) created -> running -> done
Answer: C) queued -> in_progress -> completed – Runs start as queued, move to in_progress when the agent begins processing, and end at completed when finished.
Question 4: You want an agent that can generate bar charts from CSV data uploaded by users. Which tool should you enable?
A) Azure AI Search B) Bing Search C) Code Interpreter D) OpenAPI
Answer: C) Code Interpreter – Code Interpreter can execute Python code in a sandbox, allowing it to read CSV files, process data, and generate visualizations.
Question 5: A developer stores the thread.id in their application database. What does this enable?
A) Resuming conversations across user sessions without replaying history B) Sharing the agent’s model weights across threads C) Automatically scaling the agent to handle more users D) Bypassing authentication for returning users
Answer: A) – Threads persist server-side with full message history. Storing the thread ID lets users resume conversations seamlessly across sessions.
Question 6: You need an agent to answer questions with real-time information from the internet. Which tool is correct?
A) File Search B) Code Interpreter C) Azure AI Search D) Bing Search
Answer: D) Bing Search – Bing Search (web grounding) provides live web search results, enabling agents to answer questions about current events.
Question 7: In the delegation multi-agent pattern, what is the role of the “manager” agent?
A) It executes all tasks itself B) It breaks complex tasks into sub-tasks and assigns them to specialist agents C) It only monitors agent performance metrics D) It replaces the need for human oversight
Answer: B) – In the delegation pattern, the manager agent decomposes tasks and routes them to specialized agents, then aggregates their results.
Question 8: What happens if you do NOT handle the requires_action status when using function calling?
A) The agent automatically retries with a different approach B) The run eventually expires or times out C) The function is executed automatically by the service D) The thread is deleted
Answer: B) – If you don’t submit tool outputs when the run status is requires_action, the run will eventually expire.
Question 9: When combining multiple tools on a single agent, what is a recommended best practice?
A) Enable all available tools to maximize agent capabilities B) Start with fewer tools and add more as needed C) Always use at least three tools per agent D) Disable all tools except function calling
Answer: B) – Starting with fewer tools and adding incrementally avoids confusing the agent and makes debugging easier.
Question 10: Your agent processes financial transactions. Users can request refunds over $10,000. What safety pattern should you implement?
A) Disable the agent for large amounts B) Use Code Interpreter to validate amounts C) Human-in-the-loop approval for high-value refunds D) Add a Bing Search tool to verify refund policies
Answer: C) Human-in-the-loop – For high-stakes financial decisions, requiring human approval before executing the action is the standard safety pattern.
Question 11: What is the key difference between Azure AI Search and File Search tools?
A) Azure AI Search works only with PDFs, File Search works with all formats B) File Search uses vector stores from uploaded files; Azure AI Search queries a pre-built enterprise index C) They are identical tools with different names D) Azure AI Search is free; File Search requires a premium tier
Answer: B) – File Search creates vector stores from uploaded files on-the-fly, while Azure AI Search connects to pre-existing enterprise search indexes.
Question 12: In Semantic Kernel’s AgentGroupChat, what does the TerminationStrategy control?
A) Which model each agent uses B) When the multi-agent conversation should stop C) How many tools each agent can access D) The maximum token count per message
Answer: B) – The TerminationStrategy defines the conditions under which the group chat ends, such as when a reviewer says “APPROVED” or after maximum iterations.
Question 13: An agent’s run status is failed. What should you inspect to understand the issue?
A) The thread’s creation timestamp
B) The agent’s instructions field
C) The run’s last_error property
D) The model’s temperature setting
Answer: C) – The last_error property on the run object contains details about why the run failed.
Question 14: You want an agent to call your company’s internal REST API for inventory lookups. Which tool is most appropriate?
A) Code Interpreter B) Bing Search C) OpenAPI tool or Function Calling D) File Search
Answer: C) OpenAPI tool or Function Calling – OpenAPI tools let agents call REST APIs directly from an OpenAPI spec. Function Calling works if you want to execute the API call in your own code.
Question 15: A multi-agent system enters an infinite loop where two agents keep responding to each other. What design flaw caused this?
A) The agents share the same model B) No maximum iteration limit was set in the orchestration strategy C) The agents have too few tools enabled D) The thread has too many messages
Answer: B) – Without a maximum iteration limit, agents can endlessly exchange messages. Always set maximum_iterations in your termination strategy.