How to design AI agent loops: schedules, goals, and subagents in Claude Code and Codex

06/17/26
Source: Lenny’s Newsletter. Read the original article

TL;DR Summary of Every loop type explained: heartbeats, crons, goals, and subagents

This article breaks down the four main loop types—heartbeat, cron, hook, and goal—and explains when to use each for effective automation. It outlines the five essentials every loop needs before production, such as work trees and subagents. Two live builds demonstrate practical loop applications in Claude Code and Codex, showcasing autonomous PR review and skills identification. It also highlights common pitfalls like costly goal-based loops and how to design loops using an employee onboarding framework.

Optimixed’s Overview: Understanding Loop Automation for Smarter Workflow Management

Introduction to Loop Automation

Loops represent a powerful way to automate prompts and tasks, moving beyond simple triggers to sophisticated, autonomous processes. Unlike traditional prompt automation, loops continuously manage workflows with greater flexibility, adapting to various scenarios.

The Four Loop Types Explained

  • Heartbeat loops: Continuous cycles that monitor and act regularly.
  • Cron loops: Scheduled tasks executed at set times or intervals.
  • Hook loops: Event-driven triggers responding to specific actions.
  • Goal loops: Complex loops focused on achieving target outcomes, often spawning subagents for validation.

Five Critical Components for Effective Loops

Before deploying loops in production, ensure they include:

  • Work trees: Structured task breakdowns.
  • Skills: Defined capabilities for task execution.
  • Plugins/connectors: Integration points for external systems.
  • Subagents: Autonomous units managing subtasks.
  • State tracking: Mechanisms to monitor progress and context.

Designing Loops with the “Onboarding an Employee” Framework

This mental model helps conceptualize loop design by paralleling how an employee learns and executes responsibilities, ensuring clarity, adaptability, and accountability within the automation.

Practical Loop Builds Demonstrated

  • Daily Aging PR Reviewer: A Claude Code loop that runs every morning to review and alert on aging pull requests, managing its own subagents.
  • Weekly Skills Identification: A Codex loop that continuously assesses and validates skill data by spinning up goal-based subagents in real-time.

Challenges and Warnings

Goal-based loops require careful crafting to avoid excessive token consumption without meaningful results. The article highlights two key warning signs indicating when a loop may become costly before delivering value, emphasizing the importance of thoughtful loop design and testing.