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.

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.

How would you bound and terminate an agent loop?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue practice

Next in path

Tech stacks

Skill areas

Human editorial direction

Was this drill useful?

Found this answer useful?