Agent legibility
The web is being read by machines acting for people. Agent legibility is the discipline of engineering a brand so those machines can find it, understand it, trust it enough to cite it, and act on it. This is the idea the rest of this site is built to prove.
The premise
Machine media changes who you are writing for
For thirty years the web was built for human eyes. That assumption is breaking. As AI agents take over more of the reading, researching, and buying, your site becomes less a destination a person visits and more a data source a machine ingests and acts on. Mike King calls this machine media, and the implication is blunt: the audience that decides your visibility is increasingly not human.
Agent legibility is the response. It is the engineering work of making a brand legible to that audience, so the machines reading the web on your customers' behalf can do their job with you instead of around you.
The framework
Five things an agent must be able to do
Read, cite, act, transact, measure. Each is a layer, and each is a place to win or lose.
| Layer | What it means | Where to start |
|---|---|---|
| Read | A machine can parse your pages without guessing | Structured data, llms.txt, markdown twins |
| Cite | A model trusts you enough to name you | AEO, GEO, entity consistency |
| Act | An agent can use your site, not scrape it | Model Context Protocol, WebMCP |
| Transact | An agent can buy or book for a customer | Agentic commerce |
| Measure | You can prove any of it is working | AI visibility |
Most of the market is stuck on the first two layers, and even those badly. The compounding advantage is in moving down the stack while it is still early, especially for ecommerce and service businesses, where act and transact turn directly into revenue.
Our stance
We will not preach what we do not practice
It would be absurd to sell agent legibility from a site agents cannot read. So this site is the reference implementation. Every page ships a clean structured-data layer and a markdown twin. We publish a curated llms.txt and a full content export. And we keep an honest, public scorecard of our own machine layer, including the parts that are still cheap insurance rather than proven wins. The credibility is the point. If we tell you something works, it is running here first.