Learn GenAI
Github
Contents
Map

Quiz · 16 · Agent Frameworks

66 questions from 10 pages

These are the Check Yourself questions from each page of the module, collected in course order. Each heading links back to the page the questions test. All module quizzes →

Choosing an Agent Framework

Check yourself
0 / 4 answered
  1. Which concern does an agent framework NOT remove from your responsibility?
  2. A process runs for three days with two human review steps and must survive restarts. Which layer do you need?
  3. Why build an agent loop by hand before adopting a framework?
  4. In the lab, two frameworks' agents cancelled another customer's order that three others refused. What caused it?

LangChain and LangGraph

Check yourself
0 / 4 answered
  1. What must a LangGraph graph have for interrupt() to work?
  2. A node sends an email and then calls interrupt(). What goes wrong on resume?
  3. Which LangChain 1.x piece replaces AgentExecutor?
  4. Where would you enforce 'refunds only on the customer's own orders' in a create_agent agent, other than inside the tool?

OpenAI Agents SDK

Check yourself
0 / 4 answered
  1. The triage agent hands off to billing. Who writes the reply the user sees?
  2. How does a run requiring approval survive a process restart?
  3. Why does the lab call set_tracing_disabled(True)?
  4. An input guardrail classifies every message with a second model call. What does it cost, and how does the SDK hide the latency?

Claude Agent SDK

Check yourself
0 / 4 answered
  1. What does the Claude Agent SDK provide that the Messages API Tool Runner does not?
  2. What name does the model see for tool refund_item on in-process MCP server 'shop'?
  3. Which runs first when the agent tries to run a Bash command?
  4. Why is bypassPermissions acceptable only in a sandbox?

Google ADK

Check yourself
0 / 4 answered
  1. A support agent should let the model pick a specialist who then talks to the user directly. Which mechanism?
  2. What happens when before_tool_callback returns a dict?
  3. Which state key prefix persists across all sessions of the same user?
  4. Why does ADK 2 deprecate SequentialAgent and LoopAgent?

CrewAI

Check yourself
0 / 4 answered
  1. In a hierarchical process, who assigns tasks to agents?
  2. In a Flow, what triggers a method decorated with @listen('review')?
  3. Why log the actual prompt before tuning a CrewAI agent?
  4. Why is a Flow the recommended backbone for a production CrewAI app?

Microsoft Agent Framework

Check yourself
0 / 4 answered
  1. Which two projects does Microsoft Agent Framework succeed?
  2. Where does a MAF workflow save checkpoints?
  3. Which orchestration keeps a task ledger and re-plans when progress stalls?
  4. How does a run with a tool marked approval_mode='always_require' continue?

PydanticAI

Check yourself
0 / 4 answered
  1. How does a tool get the current database connection in PydanticAI?
  2. What happens when the model's output fails validation against output_type?
  3. Why does TestModel make agent unit tests cheap?
  4. What does wrapping an agent in TemporalAgent change?

Q&A Review Bank

Check yourself
0 / 6 answered
  1. What does an agent framework add to a raw model API loop? Name five things.
  2. What does a framework hide that you need when debugging?
  3. Place these on the layer map: LangGraph, OpenAI Agents SDK, anthropic SDK, Bedrock AgentCore.
  4. When is no framework the right choice?
  5. The same model and tools ran in six frameworks in this module's lab. Where did results differ, and why?
  6. How do you keep the cost of switching frameworks low?
Check yourself
0 / 5 answered
  1. What replaced AgentExecutor in LangChain 1.x?
  2. Name four built-in LangChain middleware and what each does.
  3. What two things does interrupt() need to work?
  4. Why must side effects come after interrupt() in a node, or be idempotent?
  5. What is the LangGraph Store for?
Check yourself
0 / 4 answered
  1. Handoff vs agent.as_tool - what's the difference?
  2. How is a run requiring approval resumed after a restart?
  3. What do input guardrails cost and how is latency hidden?
  4. What do you lose running the Agents SDK on a non-OpenAI model through Chat Completions?
Check yourself
0 / 4 answered
  1. How does the Claude Agent SDK differ from the Messages API Tool Runner?
  2. How are custom tools added, and what name does the model see?
  3. In what order are permissions decided?
  4. Which hook would you use to make the agent run tests before finishing?
Check yourself
0 / 4 answered
  1. What are ADK's three ways to involve another agent?
  2. What replaced SequentialAgent, ParallelAgent and LoopAgent in ADK 2?
  3. What happens when before_tool_callback returns a dict?
  4. What do the state prefixes user:, app: and temp: mean?
Check yourself
0 / 3 answered
  1. What does a hierarchical crew add over a sequential one?
  2. In a Flow, what do @start, @listen and @router do?
  3. Why is a Flow the recommended backbone for production CrewAI apps?
Check yourself
0 / 5 answered
  1. What does Microsoft Agent Framework combine?
  2. Name MAF's five built-in orchestrations.
  3. How does PydanticAI pass a database connection to tools?
  4. How does PydanticAI pause for approval?
  5. How does PydanticAI make runs durable?

One Agent, Many Frameworks

Check yourself
0 / 3 answered
  1. Why does the lab keep the tools as plain functions in common.py rather than writing them per framework?
  2. Which approval styles let a reviewer approve tomorrow from another process?
  3. The OpenAI Agents SDK scored 12/18 and LangChain 15/18. Is LangChain the better framework?