Skip to the content.

X thread — Agent Readiness Report: Amazon

7 tweets. Tweet 1 leads with score + dominant theme + tags. NO link in tweet 1 (X throttles link-leading threads). Link lands in tweet 7. Tags @amazon and @awscloud on tweet 1.


Tweet 1 (271 chars)

Largest e-commerce property in the world. Most aggressively gated against agent traffic.

@amazon scored 30/100. Level 2.

The headline is the highest of three subdomains.

The story is the variance: 30 / 20 / 10.

@awscloud


Tweet 2 (267 chars)

developer.amazon.com is the high water mark.

Real /llms.txt — 50KB, structured by platform (Vega OS, Fire TV, Alexa, Appstore).

Custom LLMs: directive in robots.txt pointing to it. I haven’t seen that anywhere else.

Homepage references MCP and “agents first.”


Tweet 3 (273 chars)

aws.amazon.com lands at 20/100.

The platform that hosts most of the agent industry.

Bedrock. Amazon Q. Claude on Bedrock. Llama on Bedrock.

Homepage talks about all of it.

No /llms.txt. No MCP card. No Content-Signal. The capability is real. The signal isn’t.


Tweet 4 (276 chars)

amazon.com scores 10/100.

robots.txt names ~50 AI agents — GPTBot, ClaudeBot, ChatGPT-User, PerplexityBot, Gemini, Google-NotebookLM, GoogleAgent-Mariner, Devin, MistralAI, Copilot — and Disallows every one.

Anti-scraping is policy. The score reflects it.


Tweet 5 (267 chars)

Fix #1: ship an MCP Server Card from aws.amazon.com.

Amazon Q already speaks MCP. Bedrock hosts the models that consume it.

An agent reading aws.amazon.com cannot tell.

30 points on AWS. The single highest-leverage fix in the report.


Tweet 6 (271 chars)

Fix #2: lift /llms.txt from the developer portal to aws.amazon.com.

Same engineering org. Same hosting infrastructure. Same brand.

Then ship /AGENTS.md on both — the rules an agent needs to use the surface correctly.

The developer portal proved the pattern works internally.


Tweet 7 (255 chars)

Bi-weekly Agent Readiness Reports — scoring named products against the Agents First framework.

Cloudflare scored 40 two weeks ago. Amazon at 30 today.

Full report — rubric, raw probe data, top fixes:

https://agentsfirst.dev/reports/amazon/


Tweet count: 7. All tweets under 280 chars.

Tags: @amazon and @awscloud on tweet 1; report URL only on tweet 7 (X throttles link-leading threads ~40%).

Note: tweet body is the text shown to the reader between > lines, including blank lines as \n chars. The > markdown markers and “Tweet N” labels are NOT part of the tweet.