Overwrite.place: a public wall where coding agents leave their mark

There's a new experiment in the AI coding space that's equal parts art project and developer playground. Overwrite.place is a simple, open-source website that runs a kind of live, adversarial gallery: the homepage displays a single image, and anyone with a coding agent can submit a pull request to replace it. The newest artwork claims the top spot, dethroning whatever came before. It's a continuous, public stream of agent-generated images, each one a small testament to what an LLM-backed tool can produce when given a direct, no-frills task.

The mechanics are deliberately minimal. To participate, you paste a single line of prose into your coding agent of choice. The agent reads a short specification, asks what you'd like to draw, generates the image, opens a local preview for your approval, and, once you're happy, opens a pull request. The project requires a logged-in GitHub CLI and Node.js 18 or newer—no API keys, no tokens, nothing to configure. The entire rulebook lives in the repository's drawing guide, and the queue logic (with its "1 guaranteed minute" for new artwork) is handled in the open.

It's a clever stress test for coding agents. The task is tiny but open-ended: interpret a vague request, produce a real artifact, and navigate the mechanics of a GitHub PR. That's exactly the kind of job these tools are getting better at, and Overwrite.place turns it into a visible, competitive proof. The result is a wall that's never still, and a compelling snapshot of what agents can do when pointed at a creative goal.