The five layers of agent readiness
One question per layer, and the files or protocols that answer it.
In short
Agent readiness splits into five layers: find, read, act, pay and trust. Each layer asks one question and is answered by different files or protocols. A site can be strong on one layer and absent on another, so check them separately. AgentReady groups the same ground into three stages: find, read and act.
The five layers
This reference groups agent readiness into five layers. Each is a question that an agent's task depends on.
- Find: can an agent find you, and is it allowed in? robots.txt, sitemaps, Content Signals and A2A Agent Cards live here.
- Read: can it read and understand what you offer? Server-rendered HTML, llms.txt, markdown negotiation (
Accept: text/markdown) and JSON-LD address this layer. - Act: can it complete a task on your site or API? MCP, WebMCP, agents.json and OpenAPI describe actions an agent can call.
- Pay: can it pay you? x402 uses HTTP 402 so a client can pay per request and retry. ACP and UCP are agentic commerce protocols.
- Trust: can you tell which agent it is, and on whose behalf it acts? Web Bot Auth and OAuth 2.0 address identity and delegated access.
Other sources cut the same ground differently. Ora scores four layers: Discovery, Access, Usability and Payments. AgentReady names three stages: find, read and act. The standards map shows where each standard sits.
Each layer can fail on its own
A site can allow every AI crawler in robots.txt and still render its pricing only in JavaScript, which fails the read layer. A site can publish a complete llms.txt and still offer no way for an agent to submit a demo request, which fails the act layer. A single score that averages everything can hide a layer that is missing entirely.
Ora's lab found (19 Oct 2026) that, across 19 configurations of one docs site, only two stopped retrieval: hiding the answer in JavaScript and returning 403 to agent user agents. Those are a read failure and a find failure. This is a vendor claim from Ora's own experiment.
Files and protocols
Find and read are mostly files you publish and pages you render: robots.txt, sitemap.xml, llms.txt, JSON-LD. Act, pay and trust are protocols you implement, usually on an API or at the CDN: an MCP server, WebMCP tools, an x402 payment flow, Web Bot Auth verification. The first two are usually content and front-end work. The last three usually need engineering time.
One quick check per layer
- Find: fetch /robots.txt and search for
Disallow: /underUser-agent: *. - Read: fetch your pricing page with JavaScript disabled and look for the prices.
- Act: write down the three to five tasks an agent should complete.
- Pay: decide whether anything on your site should be bought per request, and record the answer.
- Trust: check whether your CDN verifies signed bots or trusts user agent strings alone.
Five checks will not make a site ready, but they show which layer to work on first.
How the layers map to AgentReady
AgentReady v2.0 has 24 live identifiers in three stages: 4 for find, 10 for read and 10 for act. It has no separate pay or trust stage. Agent payments are AR-ACT-08, a MAY. Verified identity reaches the spec through Web Bot Auth and AR-FIND-01, the MUST on not blocking agents. The AgentReady page lists the standards each identifier touches.
Sources
- RFC 9309, Robots Exclusion Protocol. https://www.rfc-editor.org/rfc/rfc9309.html, read 2026-09.
- llmstxt.org. https://llmstxt.org/, read 2026-09.
- MCP specification. https://modelcontextprotocol.io/specification/latest, read 2026-09.
- WebMCP repository. https://github.com/webmachinelearning/webmcp, read 2026-09.
- x402.org. https://www.x402.org/, read 2026-09.
- Cloudflare, Web Bot Auth docs. https://developers.cloudflare.com/bots/reference/bot-verification/web-bot-auth/, read 2026-09.
- AgentReady specification, v2.0. https://github.com/agentready-org/standard/blob/main/spec/spec.md, read 2026-11-04.
- Ora, AgentReady launch post. https://ora.ai/blog/agentready-launch, read 2026-11-04.
Last reviewed 2026-11-05