"I No Longer Write Prompts": The Shock of "Loop Engineering" Revealed by the Creator of Claude Code

A new way to work with AI

Boris Cherny, who helped build Claude Code, an AI coding tool, said in a recent interview that he has stopped typing prompts to Claude. Instead, he sets up loops. A loop is a cycle where the AI keeps working, checks its own results, and decides what to do next. Cherny said he now mostly reviews and approves the AI's work instead of giving step-by-step instructions.

Key facts

• Cherny says that since early 2026, he has not written a single line of code by hand.
• He says he uses his phone to send about 10 pull requests (requests to change code) per day and approves them.
• A post by Peter Steinberger about "loop" design reached 500,000 views in 24 hours on X.
• Andrej Karpathy, a well-known AI researcher, also described the idea with the word "Loopcraft."

Where the idea came from

For years, people mostly talked to AI by typing one request at a time. Newer tools can run tasks for longer without a person watching each step. Addy Osmani, an engineer at Google, wrote an essay called "Loop Engineering" that gave the idea a clear definition. Andrew Ng, a well-known AI educator, also said that loop engineering has become a hot topic.

Why it matters

If this approach spreads, the main job of a programmer may change. Instead of writing every instruction, people may design systems that let AI check and improve its own work. This could make software work faster, but it also raises new questions about how much humans should review and control.

What to watch

• Whether more AI companies adopt loop-based tools.
• How well AI loops work on big, complex projects.
• How companies decide who checks and approves AI work.