Architecture scenarioHard~15 minAgents & Workflows / 3 of 18

Interview question

How would you choose an agentic workflow pattern?

Chooses among routing, sequential, parallel, evaluator-optimizer, and bounded agent loops from the task's dependency structure.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

I use routing when one classifier can select a known specialist, sequential steps when later work depends on earlier output, and parallel work when independent subproblems can be combined safely. An evaluator-optimizer loop fits an output that can be checked and improved with clear stopping rules. A free agent loop is reserved for tasks where the next step genuinely depends on intermediate evidence.

Full answer below

Practice focus

Workflow decomposition, parallelism, feedback loops, and proportional orchestration.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
The workflow shape should follow task dependencies, not a fashionable orchestration diagram.
I use routing when one classifier can select a known specialist, sequential steps when later work depends on earlier output, and parallel work when independent subproblems can be combined safely. An evaluator-optimizer loop fits an output that can be checked and improved with clear stopping rules. A free agent loop is reserved for tasks where the next step genuinely depends on intermediate evidence.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
Each added model call must improve measured outcomes enough to justify latency, cost, and another failure boundary. Parallel branches need bounded fan-out and deterministic aggregation; sequential workflows need durable state if they can outlive a request. Evaluator loops require an independent useful criterion rather than one model endlessly rewriting another model's prose. I keep deterministic operations deterministic.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
I model the workflow as explicit states and allowed transitions even when the model proposes the next action. Branches carry provenance so the final answer can be traced back to evidence and tool results. Time, token, step, and retry budgets are allocated across the graph, with partial-result and cancellation behavior defined in advance. I compare the pattern with a single-call baseline and remove orchestration that does not improve important failure slices.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Using multi-agent debate without evidence it improves the task.
Parallelizing dependent actions.
Creating an unbounded evaluator loop.

Red flagsCandidateRed flagsSignals that an interviewer may treat as risky.Open details

Details

No one can draw the allowed state transitions.
Fan-out has no maximum.
Extra calls have no measured benefit.

Follow-up prompts

Quick interviewer prompts inside this drill.

Follow-up questionsCandidateFollow-up questionsLikely next questions after the main answer.Open details

Details

Treat these as the next interviewer turn. Answer one out loud, then save the gap in your practice notes.

When is evaluator-optimizer useful?
Which branches can run in parallel?
How would you represent the workflow state?

Continue the section

Next in Agents & Workflows: Curate an agent toolset

Step 4 of 18

How would you choose and describe tools for an AI agent?

Practice route

Read core answer Review checklist Follow-up prompts Track progress

Finished this drill?

Practice freely. Sign in only when you want to save completion, confidence, notes, or bookmarks.

New to Aporeon? Create account

Drill details

Human editorial direction

Aporeon is shaped by Aleksandar Tomovski, a software developer with experience on both sides of technical interviews. Content is reviewed for accuracy, natural spoken delivery, useful depth, and honest trade-offs.

About the creator Editorial standards

Was this drill useful?

Help improve the interview library.

Found this answer useful?

Say thanks with a standalone, one-time $5 contribution. No account required.

No subscription or perks.Choose another amount

Opening drill

Architecture scenarioHard~15 minAgents & Workflows / 3 of 18

Interview question

How would you choose an agentic workflow pattern?

Chooses among routing, sequential, parallel, evaluator-optimizer, and bounded agent loops from the task's dependency structure.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

Full answer below

Practice focus

Workflow decomposition, parallelism, feedback loops, and proportional orchestration.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
The workflow shape should follow task dependencies, not a fashionable orchestration diagram.
I use routing when one classifier can select a known specialist, sequential steps when later work depends on earlier output, and parallel work when independent subproblems can be combined safely. An evaluator-optimizer loop fits an output that can be checked and improved with clear stopping rules. A free agent loop is reserved for tasks where the next step genuinely depends on intermediate evidence.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
Each added model call must improve measured outcomes enough to justify latency, cost, and another failure boundary. Parallel branches need bounded fan-out and deterministic aggregation; sequential workflows need durable state if they can outlive a request. Evaluator loops require an independent useful criterion rather than one model endlessly rewriting another model's prose. I keep deterministic operations deterministic.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
I model the workflow as explicit states and allowed transitions even when the model proposes the next action. Branches carry provenance so the final answer can be traced back to evidence and tool results. Time, token, step, and retry budgets are allocated across the graph, with partial-result and cancellation behavior defined in advance. I compare the pattern with a single-call baseline and remove orchestration that does not improve important failure slices.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Using multi-agent debate without evidence it improves the task.
Parallelizing dependent actions.
Creating an unbounded evaluator loop.

Red flagsCandidateRed flagsSignals that an interviewer may treat as risky.Open details

Details

No one can draw the allowed state transitions.
Fan-out has no maximum.
Extra calls have no measured benefit.

Follow-up prompts

Quick interviewer prompts inside this drill.

Follow-up questionsCandidateFollow-up questionsLikely next questions after the main answer.Open details

Details

Treat these as the next interviewer turn. Answer one out loud, then save the gap in your practice notes.

When is evaluator-optimizer useful?
Which branches can run in parallel?
How would you represent the workflow state?

Continue the section

Next in Agents & Workflows: Curate an agent toolset

Step 4 of 18

How would you choose and describe tools for an AI agent?

How would you choose an agentic workflow pattern?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section

How would you choose an agentic workflow pattern?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section