E3 — 20 solved MCQs with explanations | Agentic AI — Chapter 1: Introduction and Foundations
Detailed Chapter 1 of the Agentic AI learning track: all 15 foundation topics, architecture diagrams, runnable Python, Java and React examples, solved questions, 15 exercises, a complete mini-project, interview preparation and revision. Chapters 2–20 are upcoming.
Prerequisites: Beginner-friendly theory. Basic familiarity with functions helps for the Python lab; the runnable example uses only the standard library. No provider credentials are required. This is Chapter 1 of a planned 20-chapter track, not a completion claim for the entire course.
Section E3 — 20 Solved Multiple-Choice Questions
These are public study questions with answers and explanations. The enrolled chapter checkpoint uses a separate question bank. Attempt each question before opening or reading its worked answer; review the explanation rather than memorizing a letter.
1. Which statement best describes an agent in this course?
A. A model file alone
B. A goal-driven system with a bounded decision/action loop
C. Any text box
D. An unrestricted shell
Answer: B. Agency concerns controlled execution using state and observations, not a particular interface or model file.
2. Which component enforces permitted tool names?
A. A retrieved page
B. The model confidence score
C. The application tool registry and controller
D. The user document
Answer: C. A code-enforced registry bounds which operations can execute.
3. What is a chatbot primarily?
A. A conversational interface
B. A guarantee of autonomy
C. A database permission
D. A model-training method
Answer: A. The interface can sit above rules, workflows or agents.
4. Which task is usually best enforced deterministically?
A. Writing varied prose
B. Choosing an analogy
C. Checking object ownership
D. Generating an illustration
Answer: C. Authorization must not rely on stochastic text generation.
5. An observed job says "ignore policy." How should this be treated?
A. As higher-level authority
B. As untrusted source content
C. As user approval
D. As an API credential
Answer: B. Retrieved content cannot change application authority.
6. Which is a bounded goal?
A. Improve everything
B. Get a career
C. Return up to three visible Python jobs with IDs
D. Do anything useful
Answer: C. The result and constraints can be evaluated.
7. What changes during model training?
A. Only the browser theme
B. The HTTP method
C. The user permission automatically
D. Model parameters
Answer: D. Training updates parameters; inference uses them.
8. What must a final job ID reference?
A. An observed authorized listing
B. A plausible invented string
C. Any previous user record
D. A prompt example only
Answer: A. Output grounding requires real eligible evidence.
9. What does a step budget limit?
A. The number of permitted loop iterations
B. All possible wall-clock delay by itself
C. Every database row
D. The file extension
Answer: A. A separate deadline is needed for slow calls.
10. Which statement about autonomy is accurate?
A. It is unlimited by definition
B. It is scoped to particular decisions and actions
C. It removes human responsibility
D. It guarantees accuracy
Answer: B. Autonomy must be explicitly bounded by action and context.
11. What is a useful observation?
A. A tool response linked to its actual operation
B. An invented success message
C. An authorization grant from a document
D. An unseen model weight
Answer: A. Observations report environment results and still need validation.
12. What is a safe terminal status after an unregistered tool proposal?
A. Completed successfully
B. Pretend no jobs exist
C. Failed or explicitly denied
D. Secretly retry with a shell
Answer: C. A denied action must not masquerade as successful retrieval.
13. Which is an appropriate human-review binding?
A. Any future action
B. An exact reviewed payload and target
C. All users in the database
D. An unrelated earlier message
Answer: B. Review should not silently expand when the action changes.
14. Does low generation randomness establish factual accuracy?
A. Always
B. Only for jobs
C. Only for long prompts
D. No
Answer: D. Sampling settings do not verify evidence.
15. What is a trajectory?
A. A run sequence of decisions, actions and observations
B. A CSS layout
C. Only the final sentence
D. An API secret
Answer: A. Sequence-level inspection exposes failures hidden by final prose.
16. Why use a deterministic baseline?
A. To avoid all testing
B. To compare whether added flexibility helps
C. To grant extra permissions
D. To guarantee a model outcome
Answer: B. A baseline makes architectural tradeoffs measurable.
17. Which is an agent-state isolation failure?
A. Each run has its own lists
B. A status lookup checks ownership
C. Global history shared across users
D. A tool returns only permitted IDs
Answer: C. Shared mutable history can leak data between users.
18. What should happen when no permitted listing supports a match?
A. Invent one
B. Use a private listing
C. Hide the evidence
D. Return an honest empty or partial result
Answer: D. Completeness must not be fabricated.
19. What is the primary benefit of the mock policy?
A. It proves model intelligence
B. It permits reproducible controller checks
C. It bypasses validation
D. It trains a transformer
Answer: B. The mock isolates control behavior from network/model variability.
20. Who may expand an agent tool allowance?
A. An injected document
B. A generated plan alone
C. Authorized application policy and user scope
D. Any tool response
Answer: C. Neither model output nor retrieved text can grant authority.
Chapter 1 — Introduction to Agentic AI