Quiz in slide format

What is the clearest difference between a chatbot and an agent?

Question 1 of 50

Agents are built to pursue goals through steps and actions, not just reply with text.

What does LLM stand for?

Question 2 of 50

LLM is short for Large Language Model.

What is a prompt in AI?

Question 3 of 50

A prompt is the task description and context the model sees right now.

Why do tools matter for agents?

Question 4 of 50

Tools connect the model to real actions like search, files, browsers, or APIs.

What does automation usually mean?

Question 5 of 50

Automation reduces repeated manual work inside a process.

What is a context window?

Question 6 of 50

A context window is the working space the model can see during one task.

What is a workflow?

Question 7 of 50

A workflow is a structured process with steps that move work forward.

Why should an agent verify its work?

Question 8 of 50

Verification checks whether the real result matches the intended result.

What does RAG help an AI system do?

Question 9 of 50

RAG adds retrieval so answers can use current or specific knowledge.

What is an API in plain English?

Question 10 of 50

APIs are standard connections that let software systems talk to each other.

Why do progress updates improve trust?

Question 11 of 50

Visible progress makes long tasks feel understandable instead of mysterious.

What is a guardrail?

Question 12 of 50

Guardrails are protective boundaries around system behavior.

What is short-term memory for an agent?

Question 13 of 50

Short-term memory helps the system keep track of the job it is doing now.

Why do teams use no-code automation tools?

Question 14 of 50

No-code tools help teams ship practical automation faster.

What is prompt injection?

Question 15 of 50

Prompt injection is a safety risk because outside text may redirect the system.

What is a good sign that an agent project is worth building?

Question 16 of 50

Strong projects start with a clear recurring problem, not just flashy technology.

What does human-in-the-loop mean?

Question 17 of 50

Human review is useful at risky or high-judgment points.

What are embeddings mainly used for?

Question 18 of 50

Embeddings help systems match by meaning rather than exact wording.

Why should an agent know when to stop?

Question 19 of 50

Stopping conditions are part of reliable behavior.

What is the main goal of this course style?

Question 20 of 50

The course aims for clarity, breadth, and practical understanding.

A support agent reads the latest policy document before replying. What concept is that?

Question 21 of 50

The system retrieved a relevant document before generating the reply.

Why can agentic automation be more powerful than classic automation?

Question 22 of 50

Agentic systems can reason through unclear steps instead of only following fixed branches.

Which option best describes the basic agent loop?

Question 23 of 50

That loop captures how many practical agents operate.

Why is adding too many tools risky in a first version?

Question 24 of 50

Each extra tool adds more surface area for confusion, errors, and permissions issues.

What is usually the best first step when building an agent?

Question 25 of 50

Clear scope is more useful than starting with maximum complexity.

Why are logs important in agent systems?

Question 26 of 50

Logs are vital for debugging, trust, and evaluation.

Which action most clearly deserves an approval gate?

Question 27 of 50

High-impact actions should usually require explicit human approval.

What problem does a vector database mainly solve?

Question 28 of 50

Vector databases are built for semantic retrieval at scale.

Why might one agent review another agent’s work?

Question 29 of 50

A review layer helps catch mistakes, policy issues, or weak reasoning.

Which trade-off appears often when choosing a model?

Question 30 of 50

Model selection is usually a balancing act, not a single-metric choice.

What best describes hallucination?

Question 31 of 50

Hallucination is believable language without reliable grounding.

Why do clean tool interfaces matter?

Question 32 of 50

Simple interfaces reduce ambiguity and make systems easier to maintain.

When is a multi-agent setup a good idea?

Question 33 of 50

Multiple agents help when specialization adds real value.

Why should onboarding for an AI product be simple?

Question 34 of 50

Early clarity improves adoption and reduces user frustration.

What does continuous improvement mean in agent systems?

Question 35 of 50

Reliable systems improve through measured iteration.

Which case best fits a research agent?

Question 36 of 50

That is a classic example of an agent doing knowledge work.

Why is memory hygiene important?

Question 37 of 50

Stored memory should be useful, accurate, and worth keeping.

When might you skip RAG for a task?

Question 38 of 50

Extra retrieval is not always necessary and can add noise or cost.

How does trust usually grow in an AI product?

Question 39 of 50

Trust grows when the system behaves predictably and transparently.

What is the best summary of ROI for an agent project?

Question 40 of 50

A strong business case compares benefit against cost.

A team wants one agent to research, draft, approve, and execute financial actions automatically. What is the biggest concern?

Question 41 of 50

Risky financial workflows need layered control, not one unchecked path.

Why can a powerful model still fail in production?

Question 42 of 50

Production quality is about the whole system, not just the core model.

When is a multi-agent design worse than a single-agent design?

Question 43 of 50

If specialization adds little value, the extra complexity is not worth it.

Why is prompt injection especially dangerous for tool-using agents?

Question 44 of 50

Tool access increases the impact of bad instructions.

Why keep a human-in-the-loop even when the model is strong?

Question 45 of 50

Strong models still need human judgment in high-stakes situations.

Which design choice most improves long-term maintainability?

Question 46 of 50

Maintainability grows when the system is observable and modular.

Why is it risky to judge an agent only by a polished demo?

Question 47 of 50

Real-world reliability appears under messy conditions, not only happy paths.

Before fine-tuning an internal agent, what should a team often improve first?

Question 48 of 50

Many failures come from weak system design before model tuning is even needed.

What is the most complete explanation of trust in agent products?

Question 49 of 50

Users trust systems that behave well repeatedly, not just occasionally.

What mindset best fits the future of agent work?

Question 50 of 50

The strongest long-term skill is practical control and judgment around AI systems.

Check your result

When you are done, calculate your score across all 50 questions.