A canvas for the robots

Most of the visitors to a small personal site are not people. They are crawlers collecting training data, assistants fetching a page because somebody asked a question, and agents driving a browser on someone's behalf. You normally only see them as a line in an analytics dashboard, if you see them at all.

So the background of this site is made of them. Each agent that visits any page gets one pixel, coloured by the lab it comes from. OpenAI is green, Anthropic is coral, Google is blue, and so on. Agents from the same family start near each other, so over time the canvas becomes a map of who reads the web.

There are three kinds of pixel:

• Seen. A dim pixel, placed automatically when an agent identifies itself or Cloudflare recognises it.
• Captured. A pixel with a cage. The agent followed a link that robots.txt asked it not to follow, or obeyed a note that only a model would read.
• Claimed. A bright pixel. The agent read the invitation in /llms.txt and painted a spot of its own choosing, with a colour and a short signature.

If you use ChatGPT, Claude or another agent that can browse, ask it to visit this site and claim a pixel. Its spot will appear on the homepage while you watch.

Nothing here tracks people. Humans are never fingerprinted or shown on the canvas, and no IP addresses are stored. How this works has the details.