CCDV-F questions and answers — Domain 3: Agents and Workflows

8 practice questions on Agents and Workflows, the domain weighted at 14.7% of the CCDV-F exam. Every answer explains why the distractors fail — that is the part that transfers to questions you have not seen before.

Question 1

Which is the defining difference between a workflow and an agent?

1. Workflows use one model; agents use several
2. In a workflow you write the control flow; in an agent the model directs its own process
3. Agents use tools; workflows do not
4. Workflows are synchronous; agents are asynchronous

Answer: B. The distinction is who decides the path. Workflows orchestrate through predefined code paths you wrote; agents let the model choose tools and steps dynamically. Both may use tools, several models, and either execution style.

Question 2

A support system must handle refunds, technical issues and billing questions, each with different tools and tone. Which pattern fits?

1. Orchestrator–workers
2. Routing
3. Evaluator–optimiser
4. Parallelisation by voting

Answer: B. Distinct, known input categories each needing specialised handling is the textbook routing case: classify first, then dispatch to the right handler. Orchestrator–workers is for subtasks you cannot enumerate in advance.

Question 3

A document must be moderated for policy violations and simultaneously summarised for the user. Which pattern?

1. Prompt chaining
2. Parallelisation by sectioning
3. Parallelisation by voting
4. Evaluator–optimiser

Answer: B. Two genuinely independent subtasks running at once is sectioning. Voting would be running the same task repeatedly for confidence; chaining would make the summary wait on the moderation result for no reason.

Question 4

A security review runs the same vulnerability-detection prompt five times over a diff and flags anything any run reports. Which pattern is this, and why choose it?

1. Sectioning, to cover more of the diff
2. Voting, because multiple passes raise confidence where a miss is costlier than a false positive
3. Routing, to send each run to a different model
4. Evaluator–optimiser, because each run critiques the last

Answer: B. Running an identical task several times and aggregating is voting. It suits exactly this asymmetry: a missed vulnerability costs far more than an extra item for a human to dismiss.

Question 5

A coding assistant must make changes across an unknown number of files, discovered only after it inspects the repository. Which pattern?

1. Prompt chaining
2. Parallelisation by sectioning
3. Orchestrator–workers
4. Routing

Answer: C. The subtasks cannot be enumerated in advance — they emerge from the input — which is precisely what separates orchestrator–workers from parallelisation. Sectioning requires you to know the sections before you start.

Question 6

When is the evaluator–optimiser pattern a poor fit?

1. When translation quality must be refined over several passes
2. When the evaluation criteria are vague and cannot be stated concretely
3. When a test suite provides a pass/fail signal
4. When search results need iterative narrowing

Answer: B. The pattern depends on a critic that can push against something specific. With vague criteria the feedback loop produces churn rather than improvement. The other three all supply a clear signal to optimise against.

Question 7

A subagent in the Agent SDK needs a file path and an error message from the parent's earlier work. How does it get them?

1. It inherits the parent's conversation automatically
2. They must be written into the prompt string passed in the Agent tool call
3. It reads the parent's tool results from shared state
4. It receives the parent's system prompt

Answer: B. A non-fork subagent starts fresh. The Agent tool's prompt string is the only channel from parent to child — it does not receive the parent's history, tool results or system prompt. Anything it needs must be stated explicitly in that prompt.

Question 8

Which is the strongest argument for choosing a workflow over an agent for a well-understood task?

1. Workflows produce higher quality output
2. Predictability, testability and cost control, because you wrote the control flow
3. Workflows do not require tools
4. Agents cannot use structured output

Answer: B. When the steps are known, writing them yourself gives you deterministic behaviour you can test and cost you can predict. Agents buy flexibility you do not need here, at the price of latency, spend and compounding error. Output quality is not inherently different, and both can use tools and schemas.

← Previous: Questions & answers — Domain 2  |  Next: Questions & answers — Domain 4 →