What is loop engineering

Designing the system that prompts an agent — and where loop.js fits in the practice

Loop engineering is the practice of designing the system that prompts an AI agent — instead of prompting it turn by turn. You define a goal, a way for the agent to find and do work, a way to verify the result, and a stop condition; the loop does the iterating. The term took hold in mid-2026 — Claude Code’s creator Boris Cherny put it as “I no longer prompt Claude; loops prompt Claude” — by then coding agents were reliable enough at long-horizon work that the scarce skill had moved from writing prompts to designing the system that writes them.

The elements of a loop

Every serious agent loop answers five questions:

Question	loop.js answer
What is the work?	a Goal — set once, judged every Round
How does it iterate?	Rounds — each starts with fresh context and reads memory from disk
Who says it’s done?	a separate, skeptical Verify agent — never the worker itself
When does it stop?	when the Loop settles (bar met, or judged impossible) — limits are guards, never goals
What re-triggers it?	any real scheduler — loop cron installs Entries into crontab, launchd, Task Scheduler, or Modal

Where loop.js stands in the practice

The simplest loops — shell scripts and minimal runners in the ralph-loop tradition — re-prompt an agent with fresh context until a script check passes or a human stops it. That shape is powerful, and loop.js keeps its core insight (fresh context per iteration, state on disk). What loop.js adds is the verdict:

• Writer ≠ grader. A self-grading agent passes its own work; the better the model, the more confidently it does so. Verify runs as a separate agent, read-only by permission — not by prompt discipline.
• Digest-first, escalate when suspicious. The judge reads the worker’s handoff digest and can escalate — inspect the work tree, run the build, read the transcript — instead of rubber-stamping a summary.
• A “not yet” must say why. The verdict’s reason is mandatory and feeds the next Round, so iteration converges instead of retrying blind.
• Impossible is an answer. A Goal that can never pass settles as a give-up instead of burning budget to the cap.
• Guards, declared. Rounds, dollars, and per-Round wall clock bound the loop; a schedule Entry declares its own lifetime (--until settled | forever, capped).

Try it

npm create @loop.js@latest my-loop
cd my-loop && npm install
export ANTHROPIC_API_KEY=sk-ant-…
npx loop run

Quickstart

From empty directory to a settled Goal.