Architecture scenarioHard~12 minAgents & Workflows / 6 of 18

Interview question

How would you bound and terminate an agent loop?

Defines success, failure, step, time, token, cost, retry, and repetition limits with useful escalation behavior.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

I define explicit terminal outcomes such as completed, needs user input, awaiting approval, blocked, failed, or budget exhausted. The runner enforces maximum turns, tool calls, elapsed time, tokens, cost, retries, and parallel branches. It stops repeated or non-progressing behavior and returns a useful status instead of asking the model to continue indefinitely.

Full answer below

Practice focus

Loop control, completion criteria, budgets, repeated-action detection, and graceful termination.

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
An agent that decides its next action still needs application-owned stopping conditions.
I define explicit terminal outcomes such as completed, needs user input, awaiting approval, blocked, failed, or budget exhausted. The runner enforces maximum turns, tool calls, elapsed time, tokens, cost, retries, and parallel branches. It stops repeated or non-progressing behavior and returns a useful status instead of asking the model to continue indefinitely.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
Progress is measured against task state, not the fluency of the latest message. Repeated tool and argument pairs, unchanged observations, oscillating plans, and consecutive failures trigger a stop or escalation. Each tool has its own timeout and retry policy inside the overall run budget. The final output is accepted only when required evidence and validations are present.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
Budgets vary by task and risk class but cannot be removed by the model. Long-running work checkpoints state before pausing, and cancellation prevents new actions while allowing in-flight outcomes to be reconciled. Termination telemetry records the reason, remaining budget, last progress marker, and action history for evaluation. A human can resume a paused run with new authority or information, but the system does not silently reset limits and repeat the same failed trajectory.

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 only a maximum-turn limit.
Letting the model declare success without outcome checks.
Resetting budgets on every retry or resume.

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

Details

The model can disable its own limits.
Repeated identical actions are not detected.
Budget exhaustion produces an ambiguous partial result.

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.

How would you detect lack of progress?
Which terminal states would you expose?
What happens after budget exhaustion?

Continue the section

Next in Agents & Workflows: Plan and replan safely

Step 7 of 18

How should an agent plan and replan during a task?

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~12 minAgents & Workflows / 6 of 18

Interview question

How would you bound and terminate an agent loop?

Defines success, failure, step, time, token, cost, retry, and repetition limits with useful escalation behavior.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

Full answer below

Practice focus

Loop control, completion criteria, budgets, repeated-action detection, and graceful termination.

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
An agent that decides its next action still needs application-owned stopping conditions.
I define explicit terminal outcomes such as completed, needs user input, awaiting approval, blocked, failed, or budget exhausted. The runner enforces maximum turns, tool calls, elapsed time, tokens, cost, retries, and parallel branches. It stops repeated or non-progressing behavior and returns a useful status instead of asking the model to continue indefinitely.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
Progress is measured against task state, not the fluency of the latest message. Repeated tool and argument pairs, unchanged observations, oscillating plans, and consecutive failures trigger a stop or escalation. Each tool has its own timeout and retry policy inside the overall run budget. The final output is accepted only when required evidence and validations are present.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
Budgets vary by task and risk class but cannot be removed by the model. Long-running work checkpoints state before pausing, and cancellation prevents new actions while allowing in-flight outcomes to be reconciled. Termination telemetry records the reason, remaining budget, last progress marker, and action history for evaluation. A human can resume a paused run with new authority or information, but the system does not silently reset limits and repeat the same failed trajectory.

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 only a maximum-turn limit.
Letting the model declare success without outcome checks.
Resetting budgets on every retry or resume.

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

Details

The model can disable its own limits.
Repeated identical actions are not detected.
Budget exhaustion produces an ambiguous partial result.

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.

How would you detect lack of progress?
Which terminal states would you expose?
What happens after budget exhaustion?

Continue the section

Next in Agents & Workflows: Plan and replan safely

Step 7 of 18

How should an agent plan and replan during a task?

How would you bound and terminate an agent loop?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section

How would you bound and terminate an agent loop?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section