AIMentor — AI Expertise Learning Platform

13 questions · pass mark 80% · 20 minute limit. The timer auto-submits at zero.

Time remaining: 20:00

Question 1easy
What is a multi-agent system?
A single agent calling one tool many timesSeveral LLM agents, each with its own role/prompt/tools, coordinating on a taskA model fine-tuned on multiple datasetsA vector database with multiple indexes
Question 2easy
What should be your DEFAULT before adopting a multi-agent system?
A hierarchical team of ten agentsA single well-designed agent with good tools (or one orchestrator + sub-agents)Fine-tuningA group chat of agents
Question 3medium
In the supervisor/worker topology, how do workers coordinate?
Workers talk directly to each other in a meshAll coordination flows through the supervisor; workers don't talk to each other (a star)Workers share one giant contextThere is no coordination
Question 4medium
When is a sequential pipeline preferable to a dynamic supervisor?
When the step order changes every runWhen the steps are a fixed, known order — no routing LLM call, no non-determinismWhen you need agents to debateNever
Question 5medium
How does a handoff/routing pattern differ from supervisor/worker?
Handoff transfers control of the conversation to a specialist; supervisor/worker integrates workers' resultsThey are identicalHandoff uses no LLMSupervisor/worker cannot use tools
Question 6hard
When is a multi-agent design genuinely justified? (Select all that apply)
The task decomposes into distinct specialties (research, writing, code, review)Context isolation per role genuinely helps keep contexts cleanDifferent roles need different tools or permissionsYou want it to look more sophisticated
Select all that apply.
Question 7hard
What is a key risk of chaining agents that feed each other?
They run too fastErrors compound as one agent's mistakes propagate to the next; add verification/critic steps for high stakesThey cannot use toolsThey share no state
Question 8hard
What is a common failure mode of free-form group-chat/network topologies?
They are too deterministicAgents ramble and burn tokens without a strong turn limit and stop conditionThey cannot include more than two agentsThey never call tools
Question 9medium
Which statement about the multi-agent framework landscape is most accurate?
Each framework implements a unique topology no other can doLangGraph, AutoGen, CrewAI, and Semantic Kernel overlap heavily; choose mostly by ecosystem fitOnly AutoGen supports multiple agentsCrewAI is the only production option
Question 10medium
A support inbox routes each message to billing, technical, or sales expertise. Which topology fits?
Sequential pipelineHandoff / routing (triage transfers the conversation to a specialist)A single generalist agent onlyDebate/critic
Question 11medium
How do the multi-agent patterns relate to LangGraph (Topic 4)?
They can't be built in LangGraphThey map onto graphs: a supervisor is a routing node and workers are nodes, with durability and HITL availableLangGraph only supports single agentsThey replace LangGraph entirely
Question 12medium
Name the star-shaped topology (two words) where one agent routes subtasks to specialists and integrates their results.
Question 13hard
In one word, what should be your default number of agents until a task proves it needs more — one or many?

0 / 13 answered

Multi-agent Patterns — Final Exam