Multi-Agent Orchestration Quiz

Answers and explanations appear as you take the quiz.

1. Instead of one monolithic agent, what does multi-agent orchestration set up?
   • Coordinating several specialized agents to collaborate on a task
   • Running one very large model with no tools
   • Replacing all agents with a single fine-tuned model
   • Removing humans from every decision permanently
2. In an orchestrator-worker design, what does the lead agent usually do?
   • Runs all tools without a task plan
   • Decomposes the task, delegates subtasks, and combines results
   • Guarantees every worker is correct
   • Replaces all termination checks
3. Which is a real cost or risk of a multi-agent workflow?
   • Guaranteed lower latency
   • No need to check evidence
   • More coordination overhead and opportunities for errors to propagate
   • Automatic agreement with ground truth
4. What role does a termination condition play in orchestration?
   • It picks the model's vocabulary
   • It encrypts the messages between agents
   • It increases the number of GPUs used
   • It prevents agents from interacting in an infinite loop
5. Which multi-agent framework does the cited paper and documentation describe?
   • AutoGen
   • LangGraph
   • CrewAI
   • Swin Transformer
6. What distinguishes centralized from decentralized orchestration?
   • Centralized uses GPUs while decentralized uses CPUs
   • In centralized control one orchestrator routes every step; in decentralized, agents hand off directly
   • Centralized has no memory while decentralized always does
   • They are identical in practice
7. In AutoGen’s UserProxyAgent example, what may happen when an executable code block arrives and no human input is provided?
   • The conversation is guaranteed to terminate successfully
   • The agent silently rewrites its model weights
   • The agent can execute the code and return the result as a message
   • The assistant discards the code block and returns no message
8. What communication model does the AutoGen paper use?
   • A single static spreadsheet
   • A one-way image classifier
   • A shared GPU cache only
   • Agents converse through customizable message-based interactions