<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>The X1Y1 Journal</title>
    <link>https://x1y1.co/journal</link>
    <description>Deep dives into engineering, AI, and modern infrastructure from X1Y1.</description>
    <language>en-us</language>
    <atom:link href="https://x1y1.co/rss.xml" rel="self" type="application/rss+xml" />
    
    <item>
      <title><![CDATA[Three MCP Python SDK Advisories, and the One an Upgrade Will Not Fix]]></title>
      <link>https://x1y1.co/journal/mcp-python-sdk-advisories-upgrade-checklist</link>
      <guid isPermaLink="true">https://x1y1.co/journal/mcp-python-sdk-advisories-upgrade-checklist</guid>
      <description><![CDATA[GitHub published three Moderate advisories for the MCP Python SDK on October 27, 2026. Two are fixed by upgrading to 1.30.0 or 2.2.0. The token audience check stays off until you turn it on, and the redirect fix can break clients that work today.]]></description>
      <pubDate>Sat, 31 Oct 2026 00:00:00 GMT</pubDate>
      <category>Security</category>
    </item>
  
    <item>
      <title><![CDATA[Jev and System One Models: Where They Actually Fit in Your AI Stack]]></title>
      <link>https://x1y1.co/journal/jev-system-one-models-where-they-fit</link>
      <guid isPermaLink="true">https://x1y1.co/journal/jev-system-one-models-where-they-fit</guid>
      <description><![CDATA[TypeSafe's Jev returns typed decisions with probabilities instead of text, at 70 to 500ms and $0.042 per million input tokens. Here is what it replaces, what it does not, and how to test it before you trust it.]]></description>
      <pubDate>Fri, 25 Sep 2026 00:00:00 GMT</pubDate>
      <category>AI</category>
    </item>
  
    <item>
      <title><![CDATA[Building RAG Systems That Don't Hallucinate]]></title>
      <link>https://x1y1.co/journal/rag-systems-that-dont-hallucinate</link>
      <guid isPermaLink="true">https://x1y1.co/journal/rag-systems-that-dont-hallucinate</guid>
      <description><![CDATA[A practical guide to retrieval-augmented generation with hybrid search, cross-encoder reranking, and citation grounding.]]></description>
      <pubDate>Tue, 28 Jul 2026 00:00:00 GMT</pubDate>
      <category>AI</category>
    </item>
  
    <item>
      <title><![CDATA[Zero-Downtime Deploys on Cloudflare Workers]]></title>
      <link>https://x1y1.co/journal/zero-downtime-deploys-cloudflare-workers</link>
      <guid isPermaLink="true">https://x1y1.co/journal/zero-downtime-deploys-cloudflare-workers</guid>
      <description><![CDATA[How we achieve atomic deployments with zero cold starts using Workers static assets and gradual rollout strategies.]]></description>
      <pubDate>Wed, 15 Jul 2026 00:00:00 GMT</pubDate>
      <category>Engineering</category>
    </item>
  
    <item>
      <title><![CDATA[MCP Servers in Production: Lessons Learned]]></title>
      <link>https://x1y1.co/journal/mcp-servers-in-production</link>
      <guid isPermaLink="true">https://x1y1.co/journal/mcp-servers-in-production</guid>
      <description><![CDATA[What breaks when Model Context Protocol servers go to production: authentication, versioning, timeouts, observability and security.]]></description>
      <pubDate>Tue, 30 Jun 2026 00:00:00 GMT</pubDate>
      <category>AI</category>
    </item>
  
    <item>
      <title><![CDATA[The Case for Edge-First Architecture]]></title>
      <link>https://x1y1.co/journal/edge-first-architecture</link>
      <guid isPermaLink="true">https://x1y1.co/journal/edge-first-architecture</guid>
      <description><![CDATA[Why we default to the edge for new projects, and the specific scenarios where you still want a traditional origin server.]]></description>
      <pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate>
      <category>Architecture</category>
    </item>
  
    <item>
      <title><![CDATA[Fine-Tuning Open Models: A Cost Analysis]]></title>
      <link>https://x1y1.co/journal/fine-tuning-open-models-cost-analysis</link>
      <guid isPermaLink="true">https://x1y1.co/journal/fine-tuning-open-models-cost-analysis</guid>
      <description><![CDATA[When does fine-tuning beat prompting? We break down training and serving costs, latency tradeoffs, and quality deltas, using published GPU prices and worked estimates.]]></description>
      <pubDate>Wed, 20 May 2026 00:00:00 GMT</pubDate>
      <category>AI</category>
    </item>
  
  </channel>
</rss>