<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Chalie Blog</title>
  <subtitle>Latest posts from the Chalie blog</subtitle>
  <link href="https://chalie.ai/feed.xml" rel="self" type="application/atom+xml" />
  <link href="https://chalie.ai/blog/" rel="alternate" type="text/html" />
  <id>https://chalie.ai/blog/</id>
  <updated>2026-05-16T00:00:00.000Z</updated>
  <author>
    <name>Chalie Team</name>
  </author>
  
  <entry>
    <title>Directing Agents, Not Writing Code: How Chalie Develops Itself</title>
    <link href="https://chalie.ai/blog/2026-05-16-directing-agents-not-writing-code/" rel="alternate" type="text/html" />
    <id>https://chalie.ai/blog/2026-05-16-directing-agents-not-writing-code/</id>
    <updated>2026-05-16T00:00:00.000Z</updated>
    <summary>An inside look at the autonomous development workflow behind Chalie — where AI agents build features, run regression gates, and improve themselves nightly while a human steers direction.</summary>
  </entry>
  
  <entry>
    <title>The Day Chalie Scheduled a Reminder to Ask About My Doctor&#39;s Appointment</title>
    <link href="https://chalie.ai/blog/2026-04-29-the-day-chalie-scheduled-a-reminder/" rel="alternate" type="text/html" />
    <id>https://chalie.ai/blog/2026-04-29-the-day-chalie-scheduled-a-reminder/</id>
    <updated>2026-04-29T00:00:00.000Z</updated>
    <summary>While testing v0.5.0, I mentioned a minor medical thing in passing. The next day I opened the scheduler and found Chalie had quietly set itself two reminders to follow up. A look at the three systems that combine into something that feels like caring.</summary>
  </entry>
  
  <entry>
    <title>The Meta-Harness, Two Weeks Later: From Interesting Experiment to Working Contributor</title>
    <link href="https://chalie.ai/blog/2026-04-21-meta-harness-v2/" rel="alternate" type="text/html" />
    <id>https://chalie.ai/blog/2026-04-21-meta-harness-v2/</id>
    <updated>2026-04-21T00:00:00.000Z</updated>
    <summary>A follow-up on our self-improvement loop. We tore out most of the orchestration, made the kernel a single skill, and watched it open and merge multiple PRs in a single afternoon.</summary>
  </entry>
  
  <entry>
    <title>Building an AI That Improves Itself: The Meta-Harness</title>
    <link href="https://chalie.ai/blog/2026-04-06-building-the-meta-harness/" rel="alternate" type="text/html" />
    <id>https://chalie.ai/blog/2026-04-06-building-the-meta-harness/</id>
    <updated>2026-04-06T00:00:00.000Z</updated>
    <summary>How we built an autonomous improvement loop that uses Claude Opus to analyze test failures, propose fixes, and create pull requests — with no human writing code.</summary>
  </entry>
  
</feed>
