OpenAI’s Harness Engineering Experiment: Zero Manually-Written Code

Press enter or click to view image in full size

OpenAI harness engineering: AI agents generating pull requests with zero human-written code, progressive disclosure architecture

Based on OpenAI’s “Harness engineering: leveraging Codex in an agent-first world” and Rohit’s harness framework

22 min read

6 days ago

Part 5 of the Harness Engineering Series

What if you could ship over a million lines of code without writing a single line yourself? Discover how OpenAI’s groundbreaking “zero manual code” experiment transformed engineers into harness architects, revolutionizing the approach to software development! 🚀 #HarnessEngineering #AIEngineering

Summary: OpenAI conducted an experiment where over one million lines of code were generated entirely by AI agents without any human-written code. This was achieved through harness engineering, which involved creating a structured repository, ensuring application legibility, and implementing mechanical architecture enforcement. The engineers shifted their role from coding to designing the environment that allows agents to produce code, resulting in a significant increase in productivity and quality of output.

Note on sources: This article is my synthesis of OpenAI’s internal Codex experiment and Rohit’s harness engineering framework. The patterns below are extrapolated from Ryan Lopopolo’s public writeup, not official OpenAI terminology.