Module 7 Quiz: Agents, Sub-Agents and AI Teams

Six questions. Click an option to check it.

---

Question 1

What distinguishes an agent from a chat assistant?

1. It uses a larger model
2. It runs in a loop, acts through tools, sees the results and continues until a stopping rule fires
3. It only works in the terminal
4. It cannot make mistakes

---

Question 2

An agent re-reads a file it already opened, forgets a constraint from the start of the session and reintroduces a bug it fixed earlier. Which problem is this?

1. Serial execution
2. Context pollution
3. Over-permissioning
4. Echo chamber

---

Question 3

What is the orchestrator's job in the orchestrator / worker pattern?

1. Write the code the workers could not finish
2. Plan, write contracts, start workers, collect and merge results, report to the human - without implementing
3. Approve and merge the pull request
4. Review each worker's transcript line by line

---

Question 4

Two workers need to agree on an interface. When and by whom should it be decided?

1. By the two workers, talking to each other
2. By the integrator, at merge time
3. By the planner or orchestrator, before the handoff, written into both contracts
4. By whichever worker finishes first

---

Question 5

Which agent should never have shell or edit permissions?

1. The implementer
2. The integrator
3. Any agent that reads untrusted input, such as an external issue or a web page
4. The test writer

---

Question 6

A reviewer agent approves every diff the implementer produces. What is the most likely cause and the fix?

1. The code is perfect; no fix needed
2. Echo chamber - same model and bias on both sides; use a different model or system prompt for the reviewer and keep the human review
3. The reviewer needs shell access to run the tests itself
4. Add a second reviewer with the same configuration