How agent-ready are your docs?
Docs for Agents
Your documentation has AI readers now.
Docs for Agents runs a five-task reading test tailored to each product with a panel of three AI models, then publishes the scorecard and the receipts.
Every report ends in a scorecard.
These cards show the top score in each letter-grade band, from A to F. The grades table ranks all 93 products.
Find the exact limits landed PARTIAL: Service, warehouse, organization, and API rate limits are all stated with exact numbers, but a knowledge-base page gives 100000 parts per service where the usage-limits and cloud-compatibility pages both give 10k.
Ships .md mirrors, llms.txt, and an MCP server, and its two webhook pages name the same parameter two different ways.
Verify an agent webhook trigger landed PARTIAL: Definition, pushing, verification, and tracking are documented, but signature formatting is missing and invocation endpoints conflict.
Find the exact limits landed PARTIAL: The pages define every requested rule, but possible extra subscription usage has no formula or exact ceiling.
Find the exact access limits landed PASS: The client page fixes order, formats, and whole-type behavior; the cloud-sources table lists six themes and 15 types.
Five first-hour jobs, each voted pass, partial, or fail.
Each product gets five first-hour jobs of its own. They start with a first working result and the exact limits. Then come one error recovery, one webhook or authentication path, and the primary SDK. Each panelist attempts the battery alone, using only what the public pages say. Every verdict needs a verbatim quote and its URL.
How the test worksThe agent infrastructure is ahead of the content that feeds it.
91 of 93 products serve llms.txt, and 87 serve a markdown mirror. 37 pass all five agent surface checks. None scores 30 of 30 on the reading test. On every product, an agent guessed or found no confident answer at least once.
How agent-ready are your docs?
Agent-ready documentation gets an AI agent from zero to a correct integration using only what the pages say. The test measures that directly: five first-hour developer jobs, three models grading independently, five agent surface checks, and a verbatim receipt behind every claim. The scorecard shows where an agent passes, guesses, or stalls.
Frequently asked questions
Is this an execution test?
No. It is a reading test: no accounts are created, no API calls run, and no code executes. A PASS means the public docs got an agent to a confident, unambiguous answer.
Which models are on the panel?
GPT 5.6 Sol, Claude Opus 5, and DeepSeek v4 Flash, run independently on an identical brief with no shared context. The panel stays fixed so scores stay comparable, and any membership change is disclosed on the affected scorecard.
Why do the panelists disagree?
Different models read differently, and disagreement is printed on the scorecard rather than averaged away. A page contradiction that splits the panel is usually the most useful finding in the report.
What is the AI Agent Readiness Score?
Fifteen reading votes at PASS 2, PARTIAL 1, FAIL 0 make 30 points, five agent surface checks at 10 points each make 50 more, and the total out of 80 becomes a US school letter grade with its percentage. The docs platform is recorded but never graded.
Can you test our docs?
Nominate the docs site on the nominate page. The test runs one product at a time, using public pages only.
The next report needs a nominee.
Nominate a docs site