Free guide · Claude & Claude Code
Loop Engineering: The Complete Guide
Stop writing prompts, start building loops. The four eras of AI engineering, Anthropic's agentic loop, and the six parts you need to build your own. No code.
What’s inside
- 1The four eras of AI engineering
- 2What a loop actually is
- 3Why loops beat one-shot prompts
- 4The six parts of a loop
- 5Build your first loop this week
- 6The patterns worth knowing by name
- 7The mistakes that break loops
You are still writing prompts. The best builders stopped months ago.
They do not sit there crafting the perfect instruction any more. They build a loop, set the goal, and walk away. The loop does the work, checks it, fixes what failed, remembers the correction, and runs again tomorrow without them.
This guide is the whole thing. What loop engineering is, why it beats writing prompts, and the six parts you need to build your own. No code required.
The four eras of AI engineering
Every era wrapped the one before it. None of them replaced the last. You still need all four.
Prompt engineering (2022). The message.
The craft of wording one input. You pack the role, the rules, the examples and the output format into a single message, and hope it lands. You tune the words and nothing else. One input goes in, one answer comes back. No memory of past chats, and no tools.
Context engineering (2024). The memory.
Andrej Karpathy named it. The model only sees a finite window, so the real job becomes choosing what goes into that window and what gets left out. You pull from docs, tools, memory and past turns, then curate, compress and drop to make it fit. Better context beats a cleverer prompt.
Harness engineering (2026). The machine.
Mitchell Hashimoto, co-founder of HashiCorp, coined this one in February 2026. The harness is everything around the model: the tools, the memory, the guardrails and the checks bolted on. His core principle is simple. Anytime you find the agent makes a mistake, you engineer a solution so it never makes that mistake again. Strip the parts away and you are back to a chatbot.
Loop engineering (now). The system.
The newest rung. Peter Steinberger and Boris Cherny put it on the map, and Addy Osmani named it as a discipline. You stop prompting at all. You design a loop that prompts the agents for you. It works, checks its own output, remembers what you teach it, and runs again. Loop engineering sits one floor above the harness. A harness is the control structure around one acting agent. A loop adds timing, helper agents, memory and repeated, self-feeding work.
"I don't prompt Claude anymore. My job is to write loops."
Boris Cherny, creator of Claude Code, Anthropic
"You shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents."
Peter Steinberger, creator of OpenClaw
That’s the first part. The rest of this guide is free too.
Charlie Hills