Why Agents Need Approval Steps

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-builders-agentic-agent-approvals

1. Which sentence best captures the main idea of 'Why Agents Need Approval Steps'?

   1. Tools and goals are unnecessary for agent design
   2. Agents and chatbots are the same thing in every way
   3. Agents should always run without limits or oversight
   4. The safest agents check with you before taking expensive or irreversible actions — sending email, making purchases, deleting files..

2. Which of the following is part of 'A real example'?

   1. Always run with no oversight
   2. You: 'Book me a flight to Denver.' Bad agent books a 1500 dollar first-class flight. Good agent: 'I am about to book the 283 dollar economy flight to Denver, leaving Friday 7am. Confirm?'
   3. Ignore cost when scaling
   4. Approve all actions automatically

3. Which of the following is part of 'Three actions that need approval'?

   1. Spending any money over a small threshold
   2. Always run with no oversight
   3. Avoid taking any actions in the world
   4. Never log what the agent did

4. Which of the following is part of 'Review date'?

   1. Use the most expensive model regardless of fit
   2. Skip every form of evaluation
   3. Reviewed in 2026. Treat fast-changing product names, prices, availability, and policy details as examples to verify before use.
   4. Run unbounded retries on any error

5. What is 'approval' in this context?

   1. A way to disable the agent's tools
   2. A trick to bypass approvals
   3. A core concept covered in Why Agents Need Approval Steps
   4. A reason to skip all logging

6. What is 'guardrail' in this context?

   1. A way to disable the agent's tools
   2. A trick to bypass approvals
   3. A core concept covered in Why Agents Need Approval Steps
   4. A reason to skip all logging

7. What is 'irreversible action' in this context?

   1. A reason to skip all logging
   2. A trick to bypass approvals
   3. A core concept covered in Why Agents Need Approval Steps
   4. A way to disable the agent's tools

8. An agent is about to send 200 emails to customers. What is the right approval pattern?

   1. Require approval on the message template and a sample, then send in batches with a kill switch
   2. Send half now and ask afterwards
   3. Auto-send because the user already said yes once
   4. Send all and let humans clean up complaints

9. What is the best response when an agent suggests an action you do not understand?

   1. Reject everything and stop using the agent
   2. Run it twice to be sure
   3. Ask the agent to explain the action and its expected effect before approving
   4. Approve it to keep things moving

10. Why is keeping a human in the loop valuable for high-stakes agent actions?

    1. It catches mistakes before they cause real-world harm
    2. It replaces the model entirely
    3. It speeds the agent up
    4. It removes the need for any logging

11. Which budget control most directly prevents runaway costs from an agent loop?

    1. A bigger model
    2. A longer context window
    3. A hard cap on steps, tokens, or dollars per task
    4. A friendly system prompt

12. What is the difference between an agent's memory and its context window?

    1. Context lasts forever; memory is cleared every minute
    2. Nothing — they are the same thing
    3. Memory is faster but less accurate than context
    4. Context is what the model sees right now; memory persists across runs

13. Why does an AI agent need 'tools' such as a browser, calendar, or code runner?

    1. Tools make the model speak more naturally
    2. Tools let the agent take actions in the world instead of only producing text
    3. Tools shrink the context window
    4. Tools replace the need for any prompts

14. Which signal best tells you an agent is stuck in a runaway loop?

    1. It finishes the task in one step
    2. It returns a short summary and stops
    3. It asks one clarifying question
    4. It keeps repeating the same tool call with no new progress

15. Before letting an agent take a destructive action, what is the safest default?

    1. Require explicit human approval for the specific action
    2. Skip approvals if the user trusts the agent
    3. Hide the action from any log
    4. Approve once and let the agent repeat forever