Prasenjit Sarkar (@stretchcloud) on X

What it claims

Prasenjit Sarkar argues in this thread that agent-readiness is following the path SEO took: first a score, then a copied checklist, then a standard backed by companies. In his view the difference is that the checklist is executable. An agent can read the spec and patch a site against it in a weekend. The standard he points to is AgentReady.org, an "Agent Experience" spec from Ora. The scanner is Cloudflare's agent scan, which reports both a numeric score and a level from Level 1 to Level 5. Sarkar cites scores rising from 20 to 67 and from 40 to 88. He states that most of the web still fails the baseline sequence of find, read, then act.

How to use it

The thread gives three AgentReady rules:

1. Put the answer in the initial HTML, not in content rendered by JavaScript.
2. Return real HTTP status codes.
3. Ship an MCP server or a CLI if agents should take actions rather than only read.

The quoted example from Nick Launches adds detail the summary leaves out. The jump from 20 to 67 moved the site from Level 1 to Level 2. So a large score gain can still leave a site at a low level. The changes were:

• an API catalog, with Link headers pointing agents to it
• a card for the MCP server, so agents can find it
• a skills index
• an AI catalog file
• DNS changes

He reports that an agent made these fixes itself, with no manual code review.

Limits

The thread does not give file names, formats or schemas for the API catalog, MCP server card, skills index or AI catalog file. It does not give the DNS records involved. It does not say how Cloudflare's numeric score maps to levels, or what Level 5 requires. The 40 to 88 example has no attribution or detail. The collaborator list comes from the thread and has not been checked against AgentReady.org. The SEO comparison is Sarkar's opinion.