<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Compound-Ai on Jeff Geiser</title>
    <link>https://jeffgeiser.dev/tags/compound-ai/</link>
    <description>Recent content in Compound-Ai on Jeff Geiser</description>
    <generator>Hugo</generator>
    <language>en-US</language>
    <lastBuildDate>Tue, 06 Oct 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://jeffgeiser.dev/tags/compound-ai/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>The Hard Part of Sovereign AI Isn&#39;t the Models</title>
      <link>https://jeffgeiser.dev/posts/the-hard-part-of-sovereign-ai-isnt-the-models/</link>
      <pubDate>Tue, 06 Oct 2026 00:00:00 +0000</pubDate>
      <guid>https://jeffgeiser.dev/posts/the-hard-part-of-sovereign-ai-isnt-the-models/</guid>
      <description>&lt;p&gt;&lt;strong&gt;Everyone&amp;rsquo;s composing small specialized models now, and everyone&amp;rsquo;s moving them onto their own hardware. What happens when these two collide.&lt;/strong&gt;&lt;/p&gt;&#xA;&lt;p&gt;&lt;em&gt;By Jeff Geiser — October 2026&lt;/em&gt;&lt;/p&gt;&#xA;&lt;hr&gt;&#xA;&lt;p&gt;Two shifts are happening.&lt;/p&gt;&#xA;&lt;p&gt;First, we stopped shipping a single model. If you&amp;rsquo;re building anything serious now, you&amp;rsquo;re composing. A small classifier to triage the request, a retrieval step, a generator, a verifier to catch the generator, etc.. Three to five specialized models wired into a pipeline, each doing their job. The research crew named it compound AI. A composed pipeline of small models beats one big model on accuracy and cost.. as long as you keep the coordination overhead in check.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
