Codex

OpenAI's coding agent grew from roughly 1 million to roughly 8 million active users in five months, folded into ChatGPT desktop and ChatGPT Work, and stands as a direct competitor to Claude Code. The more specific record is what surrounds the model.

In February 2026, OpenAI published an account of an internal method it calls harness engineering: three engineers, later seven, produced a beta product of about one million lines of code in five months, with zero lines written by hand. Codex agents opened, evaluated and merged around 1,500 pull requests, an average of 3.5 per engineer per day. The first commit in an empty repository came from Codex CLI running GPT-5. In its Agentic AI Foundation announcement, OpenAI credits Codex CLI with more than two million merged public pull requests on GitHub.

Vivek Trivedy of LangChain names the coupling: agent products like Claude Code and Codex are post-trained with their harness, so altering a tool's logic can degrade performance. Tony Seale files both under the same pattern, model plus harness, and expects the orchestration frameworks around them to fall away.

Eric S. Raymond reports feature changes, refactoring and debugging across 63 projects in C, Go, Rust, Python and shell, and says the old drift near the context limit now arrives as a red warning to clear the session. At Block, Codex sat beside goose and Claude Code as one of three tools with incompatible interfaces and context handling, the problem Berd was built to absorb.