Autonomous coding agents: A Codex example

This post was originally published on this site

Martin Fowler

In the past few weeks, multiple “autonomous background coding agents” have been released.

Supervised coding agents: Interactive chat agents that are driven and steered by a developer. Create code locally, in the IDE. Tool examples: GitHub Copilot, Windsurf, Cursor, Cline, Roo Code, Claude Code, Aider, Goose, … Autonomous background coding agents: Headless agents that you send off to work autonomously through a whole task. Code gets created in an environment spun up exclusively for that agent, and usually results in a pull request. Some of them also are runnable locally though. Tool examples: OpenAI Codex, Google Jules, Cursor background agents, Devin, …

I gave a task to OpenAI Codex and some other agents to see what I can learn. The following is a record of one particular Codex run, to help you look behind the scenes and draw your own conclusions, followed by some of my own observations.

The

To read the full article click on the 'post' link at the top.